[Significance] Rice smart unmanned farm is the core component of smart agriculture, and it is a key path to realize the modernization of rice production and promote the high-quality development of agriculture. Leveraging advanced information technologies such as the Internet of Things (IoT) and artificial intelligence (AI), these farms enable deep integration of data-driven decision making and intelligent machines. This integration creates an unmanned production system that covers the entire process from planting and managing rice crops to harvesting, greatly improving the efficiency and precision of rice cultivation. [Progress] This paper systematically sorted out the key technologies of rice smart unmanned farms in the three main links of pre-production, production and post-production, and the key technologies of pre-production mainly include the construction of high-standard farmland, unmanned nursery, land leveling, and soil nutrient testing. The construction of high-standard farmland is the foundation of the physical environment of the smart unmanned farms of rice, which provides perfect operating environment for the operation of modernized smart farm machinery through the reasonable layout of the field roads, good drainage and irrigation systems, and the scientific planting structure. Agricultural machine operation provides a perfect operating environment. The technical level of unmanned nursery directly determines the quality of rice cultivation and harvesting in the later stage, and a variety of rice seeding machines and nursery plate setting machines have been put into use. Land leveling technology can improve the growing environment of rice and increase the land utilization rate, and the current land leveling technology through digital sensing and path planning technology, which improves the operational efficiency and reduces the production cost at the same time. Soil nutrient detection technology is mainly detected by electrochemical analysis and spectral analysis, but both methods have their advantages and disadvantages, how to integrate the two methods to achieve an all-round detection of soil nutrient content is the main direction of future research. The key technologies in production mainly include rice dry direct seeding, automated transplanting, precise variable fertilization, intelligent irrigation, field weed management, and disease diagnosis. Among them, the rice dry direct seeding technology requires the planter to have high precision and stability to ensure reasonable seeding depth and density. Automated rice transplanting technology mainly includes three ways: root washing seedling machine transplanting, blanket seedling machine transplanting, and potting blanket seedling machine transplanting; at present, the incidence of problems in the automated transplanting process should be further reduced, and the quality and efficiency of rice machine transplanting should be improved. Precision variable fertilization technology is mainly composed of three key technologies: information perception, prescription decision-making and precise operation, but there are still fewer cases of unmanned farms combining the three technologies, and in the future, the main research should be on the method of constructing the whole process operation system of variable fertilization. The smart irrigation system is based on the water demand of the whole life cycle of rice to realize adaptive irrigation control, and the current smart irrigation technology can automatically adjust the irrigation strategy through real-time monitoring of soil, climate and crop growth conditions to further improve irrigation efficiency and agricultural production benefits. The field weed management and disease diagnosis technology mainly recognizes rice weeds as well as diseases through deep learning and other methods, and combines them with precision application technology for prevention and intervention. Post-production key technologies mainly include rice yield estimation, unmanned harvesting, rice storage and processing quality testing. Rice yield estimation technology is mainly used to predict yield by combining multi-source data and algorithms, but there are still problems such as the difficulty of integrating multi-source data, which requires further research. In terms of unmanned aircraft harvesting technology, China's rice combine harvester market has tended to stabilize, and the safety of the harvester's autopilot should be further improved in the future. Rice storage and processing quality detection technology mainly utilizes spectral technology and machine vision technology to detect spectra and images, and future research can combine deep learning and multimodal fusion technology to improve the machine vision system's ability and adaptability to recognize the appearance characteristics of rice. [Conclusions and Prospects] This paper reviews the researches of the construction of intelligent unmanned rice farms at home and abroad in recent years, summarizes the main difficulties faced by the key technologies of unmanned farms in practical applications, analyzes the challenges encountered in the construction of smart unmanned farms, summarizes the roles and responsibilities of the government, enterprises, scientific research institutions, cooperatives and other subjects in promoting the construction of intelligent unmanned rice farms, and puts forward relevant suggestions. It provides certain support and development ideas for the construction of intelligent unmanned rice farms in China.

[Significance] With the rapid development of robotics technology and the persistently rise of labor costs, the application of robots in facility agriculture is becoming increasingly widespread. These robots can enhance operational efficiency, reduce labor costs, and minimize human errors. However, the complexity and diversity of facility environments, including varying crop layouts and lighting conditions, impose higher demands on robot navigation. Therefore, achieving stable, accurate, and rapid navigation for robots has become a key issue. Advanced sensor technologies and algorithms have been proposed to enhance robots' adaptability and decision-making capabilities in dynamic environments. This not only elevates the automation level of agricultural production but also contributes to more intelligent agricultural management. [Progress] This paper reviews the key technologies of automatic navigation for facility agricultural robots. It details beacon localization, inertial positioning, simultaneous localization and mapping (SLAM) techniques, and sensor fusion methods used in autonomous localization and mapping. Depending on the type of sensors employed, SLAM technology could be subdivided into vision-based, laser-based and fusion systems. Fusion localization is further categorized into data-level, feature-level, and decision-level based on the types and stages of the fused information. The application of SLAM technology and fusion localization in facility agriculture has been increasingly common. Global path planning plays a crucial role in enhancing the operational efficiency and safety of facility aricultural robots. This paper discusses global path planning, classifying it into point-to-point local path planning and global traversal path planning. Furthermore, based on the number of optimization objectives, it was divided into single-objective path planning and multi-objective path planning. In regard to automatic obstacle avoidance technology for robots, the paper discusses sevelral commonly used obstacle avoidance control algorithms commonly used in facility agriculture, including artificial potential field, dynamic window approach and deep learning method. Among them, deep learning methods are often employed for perception and decision-making in obstacle avoidance scenarios. [Conclusions and Prospects] Currently, the challenges for facility agricultural robot navigation include complex scenarios with significant occlusions, cost constraints, low operational efficiency and the lack of standardized platforms and public datasets. These issues not only affect the practical application effectiveness of robots but also constrain the further advancement of the industry. To address these challenges, future research can focus on developing multi-sensor fusion technologies, applying and optimizing advanced algorithms, investigating and implementing multi-robot collaborative operations and establishing standardized and shared data platforms.

[Significance] The rapid development of artificial intelligence and automation has greatly expanded the scope of agricultural automation, with applications such as precision farming using unmanned machinery, robotic grazing in outdoor environments, and automated harvesting by orchard-picking robots. Collaborative operations among multiple agricultural robots enhance production efficiency and reduce labor costs, driving the development of smart agriculture. Multi-robot simultaneous localization and mapping (SLAM) plays a pivotal role by ensuring accurate mapping and localization, which are essential for the effective management of unmanned farms. Compared to single-robot SLAM, multi-robot systems offer several advantages, including higher localization accuracy, larger sensing ranges, faster response times, and improved real-time performance. These capabilities are particularly valuable for completing complex tasks efficiently. However, deploying multi-robot SLAM in agricultural settings presents significant challenges. Dynamic environmental factors, such as crop growth, changing weather patterns, and livestock movement, increase system uncertainty. Additionally, agricultural terrains vary from open fields to irregular greenhouses, requiring robots to adjust their localization and path-planning strategies based on environmental conditions. Communication constraints, such as unstable signals or limited transmission range, further complicate coordination between robots. These combined challenges make it difficult to implement multi-robot SLAM effectively in agricultural environments. To unlock the full potential of multi-robot SLAM in agriculture, it is essential to develop optimized solutions that address the specific technical demands of these scenarios. [Progress] Existing review studies on multi-robot SLAM mainly focus on a general technological perspective, summarizing trends in the development of multi-robot SLAM, the advantages and limitations of algorithms, universally applicable conditions, and core issues of key technologies. However, there is a lack of analysis specifically addressing multi-robot SLAM under the characteristics of complex agricultural scenarios. This study focuses on the main features and applications of multi-robot SLAM in complex agricultural scenarios. The study analyzes the advantages and limitations of multi-robot SLAM, as well as its applicability and application scenarios in agriculture, focusing on four key components: multi-sensor data fusion, collaborative localization, collaborative map building, and loopback detection. From the perspective of collaborative operations in multi-robot SLAM, the study outlines the classification of SLAM frameworks, including three main collaborative types: centralized, distributed, and hybrid. Based on this, the study summarizes the advantages and limitations of mainstream multi-robot SLAM frameworks, along with typical scenarios in robotic agricultural operations where they are applicable. Additionally, it discusses key issues faced by multi-robot SLAM in complex agricultural scenarios, such as low accuracy in mapping and localization during multi-sensor fusion, restricted communication environments during multi-robot collaborative operations, and low accuracy in relative pose estimation between robots. [Conclusions and Prospects] To enhance the applicability and efficiency of multi-robot SLAM in complex agricultural scenarios, future research needs to focus on solving these critical technological issues. Firstly, the development of enhanced data fusion algorithms will facilitate improved integration of sensor information, leading to greater accuracy and robustness of the system. Secondly, the combination of deep learning and reinforcement learning techniques is expected to empower robots to better interpret environmental patterns, adapt to dynamic changes, and make more effective real-time decisions. Thirdly, large language models will enhance human-robot interaction by enabling natural language commands, improving collaborative operations. Finally, the integration of digital twin technology will support more intelligent path planning and decision-making processes, especially in unmanned farms and livestock management systems. The convergence of digital twin technology with SLAM is projected to yield innovative solutions for intelligent perception and is likely to play a transformative role in the realm of agricultural automation. This synergy is anticipated to revolutionize the approach to agricultural tasks, enhancing their efficiency and reducing the reliance on labor.

[Objective] In the agricultural production, accurately classifying dried daylily grades is a critical task with significant economic implications. However, current target detection models face challenges such as inadequate accuracy and excessive parameters when applied to dried daylily grading, limiting their practical application and widespread use in real-world settings. To address these issues, an innovative lightweight YOLOv10-AD network model was proposed. The model aims to enhance detection accuracy by optimizing the network structure and loss functions while reducing parameters and computational costs, making it more suitable for deployment in resource-constrained agricultural production environments. [Methods] The dried daylilies selected from the Qingyang region of Gansu province as the research subject. A large number of images of dried daylilies, categorized into three grades superior, medium, and inferior, were collected using mobile phones under varying lighting conditions and backgrounds. The images were carefully annotated and augmented to build a comprehensive dataset for dried daylily grade classification. YOLOv10 was chosen as the base network, and a newly designed backbone network called AKVanillaNet was introduced. AKVanillaNet combines AKConv (adaptive kernel convolution) with VanillaNet's deep learning and shallow inference mechanisms. The second convolutional layer in VanillaNet was replaced with AKConv, and AKConv was merged with standard convolution layers at the end of the training phase to optimize the model for capturing the unique shape characteristics of dried daylilies. This innovative design not only improved detection accuracy but also significantly reduced the number of parameters and computational costs. Additionally, the DysnakeConv module was integrated into the C2f structure, replacing the Bottleneck layer with a Bottleneck-DS layer to form the new C2f-DysnakeConv module. This module enhanced the model's sensitivity to the shapes and boundaries of targets, allowing the neural network to better capture the shape information of irregular objects like dried daylilies, further improving the model's feature extraction capability. The Powerful-IOU (PIOU) loss function was also employed, which introduced a target-size-adaptive penalty factor and a gradient adjustment function. This design guided the anchor box regression along a more direct path, helping the model better fit the data and improve overall performance. [Results and Discussions] The testing results on the dried daylily grade classification dataset demonstrated that the YOLOv10-AD model achieved a mean average precision (mAP) of 85.7%. The model's parameters, computational volume, and size were 2.45 M, 6.2 GFLOPs, and 5.0 M, respectively, with a frame rate of 156 FPS. Compared to the benchmark model, YOLOv10-AD improved mAP by 5.7% and FPS by 25.8%, while reducing the number of parameters, computational volume, and model size by 9.3%, 24.4%, and 9.1%, respectively. These results indicated that YOLOv10-AD not only improved detection accuracy but also reduced the model's complexity, making it easier to deploy in real-world production environments. Furthermore, YOLOv10-AD outperformed larger models in the same series, such as YOLOv10s and YOLOv10m. Specifically, the weight, parameters, and computational volume of YOLOv10-AD were only 31.6%, 30.5%, and 25.3% of those in YOLOv10s, and 15.7%, 14.8%, and 9.8% of YOLOv10m. Despite using fewer resources, YOLOv10-AD achieved a mAP increase of 2.4% over YOLOv10s and 1.9% over YOLOv10m. These findings confirm that YOLOv10-AD maintains high detection accuracy while requiring significantly fewer resources, making it more suitable for agricultural production environments where computational capacity may be limited. The study also examined the performance of YOLOv10-AD under different lighting conditions. The results showed that YOLOv10-AD achieved an average accuracy of 92.3% in brighter environments and 78.6% in darker environments. In comparison, the YOLOv10n model achieved 88.9% and 71.0% in the same conditions, representing improvements of 3.4% and 7.6%, respectively. These findings demonstrate that YOLOv10-AD has a distinct advantage in maintaining high accuracy and confidence in grading dried daylilies across varying lighting conditions. [Conclusions] The YOLOv10-AD network model proposed significantly reduces the number of parameters and computational costs without compromising detection accuracy. This model presents a valuable technical reference for intelligent classification of dried daylily grades in agricultural production environments, particularly where resources are constrained.

[Significance] Fruit-picking robot stands as a crucial solution for achieving intelligent fruit harvesting. Significant progress has been made in developing foundational methods for picking robots, such as fruit recognition, orchard navigation, path planning for picking, and robotic arm control, the practical implementation of a seamless picking system that integrates sensing, movement, and picking capabilities still encounters substantial technical hurdles. In contrast to current picking systems, the next generation of fruit-picking robots aims to replicate the autonomous skills exhibited by human fruit pickers. This involves effectively performing ongoing tasks of perception, movement, and picking without human intervention. To tackle this challenge, this review delves into the latest research methodologies and real-world applications in this field, critically assesses the strengths and limitations of existing methods and categorizes the essential components of continuous operation into three sub-modules: local target recognition, global mapping, and operation planning. [Progress] Initially, the review explores methods for recognizing nearby fruit and obstacle targets. These methods encompass four main approaches: low-level feature fusion, high-level feature learning, RGB-D information fusion, and multi-view information fusion, respectively. Each of these approaches incorporates advanced algorithms and sensor technologies for cluttered orchard environments. For example, low-level feature fusion utilizes basic attributes such as color, shapes and texture to distinguish fruits from backgrounds, while high-level feature learning employs more complex models like convolutional neural networks to interpret the contextual relationships within the data. RGB-D information fusion brings depth perception into the mix, allowing robots to gauge the distance to each fruit accurately. Multi-view information fusion tackles the issue of occlusions by combining data from multiple cameras and sensors around the robot, providing a more comprehensive view of the environment and enabling more reliable sensing. Subsequently, the review shifts focus to orchard mapping and scene comprehension on a broader scale. It points out that current mapping methods, while effective, still struggle with dynamic changes in the orchard, such as variations of fruits and light conditions. Improved adaptation techniques, possibly through machine learning models that can learn and adjust to different environmental conditions, are suggested as a way forward. Building upon the foundation of local and global perception, the review investigates strategies for planning and controlling autonomous behaviors. This includes not only the latest advancements in devising movement paths for robot mobility but also adaptive strategies that allow robots to react to unexpected obstacles or changes within the whole environment. Enhanced strategies for effective fruit picking using the Eye-in-Hand system involve the development of more dexterous robotic hands and improved algorithms for precisely predicting the optimal picking point of each fruit. The review also identifies a crucial need for further advancements in the dynamic behavior and autonomy of these technologies, emphasizing the importance of continuous learning and adaptive control systems to improve operational efficiency in diverse orchard environments. [Conclusions and Prospects] The review underscores the critical importance of coordinating perception, movement, and picking modules to facilitate the transition from a basic functional prototype to a practical machine. Moreover, it emphasizes the necessity of enhancing the robustness and stability of core algorithms governing perception, planning, and control, while ensuring their seamless coordination which is a central challenge that emerges. Additionally, the review raises unresolved questions regarding the application of picking robots and outlines future trends, include deeper integration of stereo vision and deep learning, enhanced global vision sampling, and the establishment of standardized evaluation criteria for overall operational performance. The paper can provide references for the eventual development of robust, autonomous, and commercially viable picking robots in the future.

[Objective] The phenomenon of hoof slipping occurs during the walking process of cows, which indicates the deterioration of the farming environment and a decline in the cows' locomotor function. Slippery grounds can lead to injuries in cows, resulting in unnecessary economic losses for farmers. To achieve automatically recognizing and detecting slippery hoof postures during walking, the study focuses on the localization and analysis of key body points of cows based on deep learning methods. Motion curves of the key body points were analyzed, and features were extracted. The effectiveness of the extracted features was verified using a decision tree classification algorithm, with the aim of achieving automatic detection of slippery hoof postures in cows. [Method] An improved localization method for the key body points of cows, specifically the head and four hooves, was proposed based on the DeepLabCut model. Ten networks, including ResNet series, MobileNet-V2 series, and EfficientNet series, were selected to respectively replace the backbone network structure of DeepLabCut for model training. The root mean square error(RMSE), model size, FPS, and other indicators were chosen, and after comprehensive consideration, the optimal backbone network structure was selected as the pre-improved network. A network structure that fused the convolutional block attention module (CBAM) attention mechanism with ResNet-50 was proposed. A lightweight attention module, CBAM, was introduced to improve the ResNet-50 network structure. To enhance the model's generalization ability and robustness, the CBAM attention mechanism was embedded into the first convolution layer and the last convolution layer of the ResNet-50 network structure. Videos of cows with slippery hooves walking in profile were predicted for key body points using the improved DeepLabCut model, and the obtained key point coordinates were used to plot the motion curves of the cows' key body points. Based on the motion curves of the cows' key body points, the feature parameter Feature1 for detecting slippery hooves was extracted, which represented the local peak values of the derivative of the motion curves of the cows' four hooves. The feature parameter Feature2 for predicting slippery hoof distances was extracted, specifically the minimum local peak points of the derivative curve of the hooves, along with the local minimum points to the left and right of these peaks. The effectiveness of the extracted Feature1 feature parameters was verified using a decision tree classification model. Slippery hoof feature parameters Feature1 for each hoof were extracted, and the standard deviation of Feature1 was calculated for each hoof. Ultimately, a set of four standard deviations for each cow was extracted as input parameters for the classification model. The classification performance was evaluated using four common objective metrics, including accuracy, precision, recall, and F₁-Score. The prediction accuracy for slippery hoof distances was assessed using RMSE as the evaluation metric. [Results and Discussion] After all ten models reached convergence, the loss values ranked from smallest to largest were found in the EfficientNet series, ResNet series, and MobileNet-V2 series, respectively. Among them, ResNet-50 exhibited the best localization accuracy in both the training set and validation set, with RMSE values of only 2.69 pixels and 3.31 pixels, respectively. The MobileNet series had the fastest inference speed, reaching 48 f/s, while the inference speeds of the ResNet series and MobileNet series were comparable, with ResNet series performing slightly better than MobileNet series. Considering the above factors, ResNet-50 was ultimately selected as the backbone network for further improvements on DeepLabCut. Compared to the original ResNet-50 network, the ResNet-50 network improved by integrating the CBAM module showed a significant enhancement in localization accuracy. The accuracy of the improved network increased by 3.7% in the training set and by 9.7% in the validation set. The RMSE between the predicted body key points and manually labeled points was only 2.99 pixels, with localization results for the right hind hoof, right front hoof, left hind hoof, left front hoof, and head improved by 12.1%, 44.9%, 0.04%, 48.2%, and 39.7%, respectively. To validate the advancement of the improved model, a comparison was made with the mainstream key point localization model, YOLOv8s-pose, which showed that the RMSE was reduced by 1.06 pixels compared to YOLOv8s-pose. This indicated that the ResNet-50 network integrated with the CBAM attention mechanism possessed superior localization accuracy. In the verification of the cow slippery hoof detection classification model, a 10-fold cross-validation was conducted to evaluate the performance of the cow slippery hoof classification model, resulting in average values of accuracy, precision, recall, and F₁-Score at 90.42%, 0.943, 0.949, and 0.941, respectively. The error in the calculated slippery hoof distance of the cows, using the slippery hoof distance feature parameter Feature2, compared to the manually calibrated slippery hoof distance was found to be 1.363 pixels. [Conclusion] The ResNet-50 network model improved by integrating the CBAM module showed a high accuracy in the localization of key body points of cows. The cow slippery hoof judgment model and the cow slippery hoof distance prediction model, based on the extracted feature parameters for slippery hoof judgment and slippery hoof distance detection, both exhibited small errors when compared to manual detection results. This indicated that the proposed enhanced deeplabcut model obtained good accuracy and could provide technical support for the automatic detection of slippery hooves in cows.

[Significance] Building the agricultural new quality productivity is of great significance. It is the advanced quality productivity which realizes the transformation, upgrading, and deep integration of substantive, penetrating, operational, and media factors, and has outstanding characteristics such as intelligence, greenness, integration, and organization. As a new technology revolution in the field of agriculture, smart agricultural technology transforms agricultural production mode by integrating agricultural biotechnology, agricultural information technology, and smart agricultural machinery and equipment, with information and knowledge as important core elements. The inherent characteristics of "high-tech, high-efficiency, high-quality, and sustainable" in agricultural new quality productivity are fully reflected in the practice of smart agricultural technology innovation. And it has become an important core and engine for promoting the agricultural new quality productivity. [Progress] Through literature review and theoretical analysis, this article conducts a systematic study on the practical foundation, internal logic, and problem challenges of smart agricultural technology innovation leading the development of agricultural new quality productivity. The conclusions show that: (1) At present, the global innovation capability of smart agriculture technology is constantly enhancing, and significant technology breakthroughs have been made in fields such as smart breeding, agricultural information perception, agricultural big data and artificial intelligence, smart agricultural machinery and equipment, providing practical foundation support for leading the development of agricultural new quality productivity. Among them, the smart breeding of 'Phenotype+Genotype+Environmental type' has entered the fast lane, the technology system for sensing agricultural sky, air, and land information is gradually maturing, the research and exploration on agricultural big data and intelligent decision-making technology continue to advance, and the creation of smart agricultural machinery and equipment for different fields has achieved fruitful results; (2) Smart agricultural technology innovation provides basic resources for the development of agricultural new quality productivity through empowering agricultural factor innovation, provides sustainable driving force for the development of agricultural new quality productivity through empowering agricultural technology innovation, provides practical paradigms for the development of agricultural new quality productivity through empowering agricultural scenario innovation, provides intellectual support for the development of agricultural new quality productivity through empowering agricultural entity innovation, and provides important guidelines for the development of agricultural new quality productivity through empowering agricultural value innovation; (3) Compared to the development requirements of agricultural new quality productivity in China and the advanced level of international smart agriculture technology, China's smart agriculture technology innovation is generally in the initial stage of multi-point breakthroughs, system integration, and commercial application. It still faces major challenges such as an incomplete policy system for technology innovation, key technologies with bottlenecks, blockages and breakpoints, difficulties in the transformation and implementation of technology achievements, and incomplete support systems for technology innovation. [Conclusions and Prospects] Regarding the issue of technology innovation in smart agriculture, this article proposes the 'Four Highs' path of smart agriculture technology innovation to fill the gaps in smart agriculture technology innovation and accelerate the formation of agricultural new quality productivity in China. The "Four Highs" path specifically includes the construction of high-energy smart agricultural technology innovation platforms, the breakthroughs in high-precision and cutting-edge smart agricultural technology products, the creation of high-level smart agricultural application scenarios, and the cultivation of high-level smart agricultural innovation talents. Finally, this article proposes four strategic suggestions such as deepening the understanding of smart agriculture technology innovation and agricultural new quality productivity, optimizing the supply of smart agriculture technology innovation policies, building a national smart agriculture innovation development pilot zone, and improving the smart agriculture technology innovation ecosystem.

[Objective] In the era of digital agriculture, real-time monitoring and predictive modeling of crop growth are paramount, especially in autonomous farming systems. Traditional crop growth models, often constrained by their reliance on static, rule-based methods, fail to capture the dynamic and multifactorial nature of vegetable crop growth. This research tried to address these challenges by leveraging the advanced reasoning capabilities of pre-trained large language models (LLMs) to simulate and predict vegetable crop growth with accuracy and reliability. Modeling the growth of vegetable crops within these platforms has historically been hindered by the complex interactions among biotic and abiotic factors. [Methods] The methodology was structured in several distinct phases. Initially, a comprehensive dataset was curated to include extensive information on vegetable crop growth cycles, environmental conditions, and management practices. This dataset incorporates continuous data streams such as soil moisture, nutrient levels, climate variables, pest occurrence, and historical growth records. By combining these data sources, the study ensured that the model was well-equipped to understand and infer the complex interdependencies inherent in crop growth processes. Then, advanced techniques was emploied for pre-training and fine-tuning LLMs to adapt them to the domain-specific requirements of vegetable crop modeling. A staged intelligent agent ensemble was designed to work within the digital twin platform, consisting of a central managerial agent and multiple stage-specific agents. The managerial agent was responsible for identifying transitions between distinct growth stages of the crops, while the stage-specific agents were tailored to handle the unique characteristics of each growth phase. This modular architecture enhanced the model's adaptability and precision, ensuring that each phase of growth received specialized attention and analysis. [Results and Discussions] The experimental validation of this method was conducted in a controlled agricultural setting at the Xiaotangshan Modern Agricultural Demonstration Park in Beijing. Cabbage (Zhonggan 21) was selected as the test crop due to its significance in agricultural production and the availability of comprehensive historical growth data. Over five years, the dataset collected included 4 300 detailed records, documenting parameters such as plant height, leaf count, soil conditions, irrigation schedules, fertilization practices, and pest management interventions. This dataset was used to train the LLM-based system and evaluate its performance using ten-fold cross-validation. The results of the experiments demonstrating the efficacy of the proposed system in addressing the complexities of vegetable crop growth modeling. The LLM-based model achieved 98% accuracy in predicting crop growth degrees and a 99.7% accuracy in identifying growth stages. These metrics significantly outperform traditional machine learning approaches, including long short-term memory (LSTM), XGBoost, and LightGBM models. The superior performance of the LLM-based system highlights its ability to reason over heterogeneous data inputs and make precise predictions, setting a new benchmark for crop modeling technologies. Beyond accuracy, the LLM-powered system also excels in its ability to simulate growth trajectories over extended periods, enabling farmers and agricultural managers to anticipate potential challenges and make proactive decisions. For example, by integrating real-time sensor data with historical patterns, the system can predict how changes in irrigation or fertilization practices will impact crop health and yield. This predictive capability is invaluable for optimizing resource allocation and mitigating risks associated with climate variability and pest outbreaks. [Conclusions] The study emphasizes the importance of high-quality data in achieving reliable and generalizable models. The comprehensive dataset used in this research not only captures the nuances of cabbage growth but also provides a blueprint for extending the model to other crops. In conclusion, this research demonstrates the transformative potential of combining large language models with digital twin technology for vegetable crop growth modeling. By addressing the limitations of traditional modeling approaches and harnessing the advanced reasoning capabilities of LLMs, the proposed system sets a new standard for precision agriculture. Several avenues also are proposed for future work, including expanding the dataset, refining the model architecture, and developing multi-crop and multi-region capabilities.

Objective The accuracy of identifying litchi pests is crucial for implementing effective control strategies and promoting sustainable agricultural development. However, the current detection of litchi pests is characterized by a high percentage of small targets, which makes target detection models challenging in terms of accuracy and parameter count, thus limiting their application in real-world production environments. To improve the identification efficiency of litchi pests, a lightweight target detection model YOLO-LP (YOLO-Litchi Pests) based on YOLOv10n was proposed. The model aimed to enhance the detection accuracy of small litchi pest targets in multiple scenarios by optimizing the network structure and loss function, while also reducing the number of parameters and computational costs. Methods Two classes of litchi insect pests (Cocoon and Gall) images were collected as datasets for modeling in natural scenarios (sunny, cloudy, post-rain) and laboratory environments. The original data were expanded through random scaling, random panning, random brightness adjustments, random contrast variations, and Gaussian blurring to balance the category samples and enhance the robustness of the model, generating a richer dataset named the CG dataset (Cocoon and Gall dataset). The YOLO-LP model was constructed after the following three improvements. Specifically, the C2f module of the backbone network (Backbone) in YOLOv10n was optimized and the C2f_GLSA module was constructed using the global-to-local spatial aggregation (GLSA) module to focus on small targets and enhance the differentiation between the targets and the backgrounds, while simultaneously reducing the number of parameters and computation. A frequency-aware feature fusion module (FreqFusion) was introduced into the neck network (Neck) of YOLOv10n and a frequency-aware path aggregation network (FreqPANet) was designed to reduce the complexity of the model and address the problem of fuzzy and shifted target boundaries. The SCYLLA-IoU (SIoU) loss function replaced the Complete-IoU (CIoU) loss function from the baseline model to optimize the target localization accuracy and accelerate the convergence of the training process. Results and Discussions YOLO-LP achieved 90.9%, 62.2%, and 59.5% for AP₅₀, AP_50:95, and AP-Small_50:95 in the CG dataset, respectively, and 1.9%, 1.0%, and 1.2% higher than the baseline model. The number of parameters and the computational costs were reduced by 13% and 17%, respectively. These results suggested that YOLO-LP had a high accuracy and lightweight design. Comparison experiments with different attention mechanisms validated the effectiveness of the GLSA module. After the GLSA module was added to the baseline model, AP₅₀, AP_50:95, and AP-Small_50:95 achieved the highest performance in the CG dataset, reaching 90.4%, 62.0%, and 59.5%, respectively. Experiment results comparing different loss functions showed that the SIoU loss function provided better fitting and convergence speed in the CG dataset. Ablation test results revealed that the validity of each model improvement and the detection performance of any combination of the three improvements was significantly better than the baseline model in the YOLO-LP model. The performance of the models was optimal when all three improvements were applied simultaneously. Compared to several mainstream models, YOLO-LP exhibited the best overall performance, with a model size of only 5.1 MB, 1.97 million parameters (Params), and a computational volume of 5.4 GFLOPs. Compared to the baseline model, the detection of the YOLO-LP performance was significantly improved across four multiple scenarios. In the sunny day scenario, AP₅₀, AP_50:95, and AP-Small_50:95 increased by 1.9%, 1.0 %, and 2.0 %, respectively. In the cloudy day scenario, AP₅₀, AP_50:95, and AP-Small_50:95 increased by 2.5%, 1.3%, and 1.3%, respectively. In the post-rain scenario, AP₅₀, AP_50:95, and AP-Small_50:95 increased by 2.0%, 2.4%, and 2.4%, respectively. In the laboratory scenario, only AP₅₀ increased by 0.7% over the baseline model. These findings indicated that YOLO-LP achieved higher accuracy and robustness in multi-scenario small target detection of litchi pests. Conclusions The proposed YOLO-LP model could improve detection accuracy and effectively reduce the number of parameters and computational costs. It performed well in small target detection of litchi pests and demonstrated strong robustness across different scenarios. These improvements made the model more suitable for deployment on resource-constrained mobile and edge devices. The model provided a valuable technical reference for small target detection of litchi pests in various scenarios.

[Objective] As one of China's most important agricultural products, apples hold a significant position in cultivation area and yield. However, during the growth process, apples are prone to various diseases that not only affect the quality of the fruit but also significantly reduce the yield, impacting farmers' economic benefits and the stability of market supply. To reduce the incidence of apple diseases and increase fruit yield, developing efficient and fast apple leaf disease detection technology is of great significance. An improved YOLOv8 algorithm was proposed to identify the leaf diseases that occurred during the growth of apples. [Methods] YOLOv8n model was selected to detect various leaf diseases such as brown rot, rust, apple scab, and sooty blotch that apples might encounter during growth. SPD-Conv was introduced to replace the original convolutional layers to retain fine-grained information and reduce model parameters and computational costs, thereby improving the accuracy of disease detection. The multi-scale dilated attention (MSDA) attention mechanism was added at appropriate positions in the Neck layer to enhance the model's feature representation capability, which allowed the model to learn the receptive field dynamically and adaptively focus on the most representative regions and features in the image, thereby enhancing the ability to extract disease-related features. Finally, inspired by the RepVGG architecture, the original detection head was optimized to achieve a separation of detection and inference architecture, which not only accelerated the model's inference speed but also enhanced feature learning capability. Additionally, a dataset of apple leaf diseases containing the aforementioned diseases was constructed, and experiments were conducted. [Results and Discussions] Compared to the original model, the improved model showed significant improvements in various performance metrics. The mAP50 and mAP50:95 achieved 88.2% and 37.0% respectively, which were 2.7% and 1.3% higher than the original model. In terms of precision and recall, the improved model increased to 83.1% and 80.2%, respectively, representing an improvement of 0.9% and 1.1% over the original model. Additionally, the size of the improved model was only 7.8 MB, and the computational cost was reduced by 0.1 G FLOPs. The impact of the MSDA placement on model performance was analyzed by adding it at different positions in the Neck layer, and relevant experiments were designed to verify this. The experimental results showed that adding MSDA at the small target layer in the Neck layer achieved the best effect, not only improving model performance but also maintaining low computational cost and model size, providing important references for the optimization of the MSDA mechanism. To further verify the effectiveness of the improved model, various mainstream models such as YOLOv7-tiny, YOLOv9-c, RetinaNet, and Faster-RCNN were compared with the propoed model. The experimental results showed that the improved model outperformed these models by 1.4%, 1.3%, 7.8%, and 11.6% in mAP50, 2.8%, 0.2%, 3.4%, and 5.6% in mAP50:95. Moreover, the improved model showed significant advantages in terms of floating-point operations, model size, and parameter count, with a parameter count of only 3.7 MB, making it more suitable for deployment on hardware-constrained devices such as drones. In addition, to assess the model's generalization ability, a stratified sampling method was used, selecting 20% of the images from the dataset as the test set. The results showed that the improved model could maintain a high detection accuracy in complex and variable scenes, with mAP50 and mAP50:95 increasing by 1.7% and 1.2%, respectively, compared to the original model. Considering the differences in the number of samples for each disease in the dataset, a class balance experiment was also designed. Synthetic samples were generated using oversampling techniques to increase the number of minority-class samples. The experimental results showed that the class-balanced dataset significantly improved the model's detection performance, with overall accuracy increasing from 83.1% to 85.8%, recall from 80.2% to 83.6%, mAP50 from 88.2% to 88.9%, and mAP50:95 from 37.0% to 39.4%. The class-balanced dataset significantly enhanced the model's performance in detecting minority diseases, thereby improving the overall performance of the model. [Conclusions] The improved model demonstrated significant advantages in apple leaf disease detection. By introducing SPD-Conv and MSDA attention mechanisms, the model achieved noticeable improvements in both precision and recall while effectively reducing computational costs, leading to more efficient detection capabilities. The improved model could provide continuous health monitoring throughout the apple growth process and offer robust data support for farmers' scientific decision-making before fruit harvesting.

[Objective] The number of effective tillers per plant is one of the important agronomic traits affecting rice yield. In order to solve the problems of high cost and low accuracy of effective tiller detection caused by dense tillers, mutual occlusion and ineffective tillers in rice, a method for dividing effective tillers and ineffective tillers in rice was proposed. Combined with the deep learning model, a high-throughput and low-cost mobile phone App for effective tiller detection in rice was developed to solve the practical problems of effective tiller investigation in rice under field conditions. [Methods] The investigations of rice tillering showed that the number of effective tillers of rice was often higher than that of ineffective tillers. Based on the difference in growth height between effective and ineffective tillers of rice, a new method for distinguishing effective tillers from ineffective tillers was proposed. A fixed height position of rice plants was selected to divide effective tillers from ineffective tillers, and rice was harvested at this position. After harvesting, cross-sectional images of rice tillering stems were taken using a mobile phone, and the stems were detected and counted by the YOLOv8 model. Only the cross-section of the stem was identified during detection, while the cross-section of the panicle was not identified. The number of effective tillers of rice was determined by the number of detected stems. In order to meet the needs of field work, a mobile phone App for effective tiller detection of rice was developed for real-time detection. GhostNet was used to lighten the YOLOv8 model. Ghost Bottle-Neck was integrated into C2f to replace the original BottleNeck to form C2f-Ghost module, and then the ordinary convolution in the network was replaced by Ghost convolution to reduce the complexity of the model. Based on the lightweight Ghost-YOLOv8 model, a mobile App for effective tiller detection of rice was designed and constructed using the Android Studio development platform and intranet penetration counting. [Results and Discussions] The results of field experiments showed that there were differences in the growth height of effective tillers and ineffective tillers of rice. The range of 52 % to 55 % of the total plant height of rice plants was selected for harvesting, and the number of stems was counted as the number of effective tillers per plant. The range was used as the division standard of effective tillers and ineffective tillers of rice. The accuracy and recall rate of effective tillers counting exceeded 99%, indicating that the standard was accurate and comprehensive in guiding effective tillers counting. Using the GhostNet lightweight YOLOv8 model, the parameter quantity of the lightweight Ghost-YOLOv8 model was reduced by 43%, the FPS was increased by 3.9, the accuracy rate was 0.988, the recall rate was 0.980, and the mAP was 0.994. The model still maintains excellent performance while light weighting. Based on the lightweight Ghost-YOLOv8 model, a mobile phone App for detecting effective tillers of rice was developed. The App was tested on 100 cross-sectional images of rice stems collected under the classification criteria established in this study. Compared with the results of manual counting of effective tillers per plant, the accuracy of the App's prediction results was 99.61%, the recall rate was 98.76%, and the coefficient of determination was 0.985 9, indicating the reliability of the App and the established standards in detecting effective tillers of rice. [Conclusions] Through the lightweight Ghost-YOLOv8 model, the number of stems in the cross-sectional images of stems collected under the standard was detected to obtain the effective tiller number of rice. An Android-side rice effective tillering detection App was developed, which can meet the field investigation of rice effective tillering, help breeders to collect data efficiently, and provide a basis for field prediction of rice yield. Further research could supplement the cross-sectional image dataset of multiple rice stems to enable simultaneous measurement of effective tillers across multiple rice plants and improve work efficiency. Further optimization and enhancement of the App's functionality is necessary to provide more tiller-related traits, such as tiller angle.

[Objective] As agriculture increasingly relies on technological innovations to boost productivity and ensure sustainability, farmers need efficient and accurate tools to aid their decision-making processes. A key challenge in this context is the retrieval of specialized agricultural knowledge, which can be complex and diverse in nature. Traditional agricultural knowledge retrieval systems have often been limited by the modalities they utilize (e.g., text or images alone), which restricts their effectiveness in addressing the wide range of queries farmers face. To address this challenge, a specialized multimodal question-answering system tailored for cabbage cultivation was proposed. The system, named Agri-QA Net, integrates multimodal data to enhance the accuracy and applicability of agricultural knowledge retrieval. By incorporating diverse data modalities, Agri-QA Net aims to provide a holistic approach to agricultural knowledge retrieval, enabling farmers to interact with the system using multiple types of input, ranging from spoken queries to images of crop conditions. By doing so, it helps address the complexity of real-world agricultural environments and improves the accessibility of relevant information. [Methods] The architecture of Agri-QA Net was built upon the integration of multiple data modalities, including textual, auditory, and visual data. This multifaceted approach enables the system to develop a comprehensive understanding of agricultural knowledge, allowed the system to learn from a wide array of sources, enhancing its robustness and generalizability. The system incorporated state-of-the-art deep learning models, each designed to handle one specific type of data. Bidirectional Encoder Representations from Transformers (BERT)'s bidirectional attention mechanism allowed the model to understand the context of each word in a given sentence, significantly improving its ability to comprehend complex agricultural terminology and specialized concepts. The system also incorporated acoustic models for processing audio inputs. These models analyzed the spoken queries from farmers, allowing the system to understand natural language inputs even in noisy, non-ideal environments, which was a common challenge in real-world agricultural settings. Additionally, convolutional neural networks (CNNs) were employed to process images from various stages of cabbage growth. CNNs were highly effective in capturing spatial hierarchies in images, making them well-suited for tasks such as identifying pests, diseases, or growth abnormalities in cabbage crops. These features were subsequently fused in a Transformer-based fusion layer, which served as the core of the Agri-QA Net architecture. The fusion process ensured that each modality—text, audio, and image—contributes effectively to the final model's understanding of a given query. This allowed the system to provide more nuanced answers to complex agricultural questions, such as identifying specific crop diseases or determining the optimal irrigation schedules for cabbage crops. In addition to the fusion layer, cross-modal attention mechanisms and domain-adaptive techniques were incorporated to refine the model's ability to understand and apply specialized agricultural knowledge. The cross-modal attention mechanism facilitated dynamic interactions between the text, audio, and image data, ensuring that the model paid attention to the most relevant features from each modality. Domain-adaptive techniques further enhanced the system's performance by tailoring it to specific agricultural contexts, such as cabbage farming, pest control, or irrigation management. [Results and Discussions] The experimental evaluations demonstrated that Agri-QA Net outperforms traditional single-modal or simple multimodal models in agricultural knowledge tasks. With the support of multimodal inputs, the system achieved an accuracy rate of 89.5%, a precision rate of 87.9%, a recall rate of 91.3%, and an F₁-Score of 89.6%, all of which are significantly higher than those of single-modality models. The integration of multimodal data significantly enhanced the system's capacity to understand complex agricultural queries, providing more precise and context-aware answers. The addition of cross-modal attention mechanisms enabled for more nuanced and dynamic interaction between the text, audio, and image data, which in turn improved the model's understanding of ambiguous or context-dependent queries, such as disease diagnosis or crop management. Furthermore, the domain-adaptive technique enabled the system to focus on specific agricultural terminology and concepts, thereby enhancing its performance in specialized tasks like cabbage cultivation and pest control. The case studies presented further validated the system's ability to assist farmers by providing actionable, domain-specific answers to questions, demonstrating its practical application in real-world agricultural scenarios. [Conclusions] The proposed Agri-QA Net framework is an effective solution for addressing agricultural knowledge questions, especially in the domain of cabbage cultivation. By integrating multimodal data and leveraging advanced deep learning techniques, the system demonstrates a high level of accuracy and adaptability. This study not only highlights the potential of multimodal fusion in agriculture but also paves the way for future developments in intelligent systems designed to support precision farming. Further work will focus on enhancing the model's performance by expanding the dataset to include more diverse agricultural scenarios, refining the handling of dialectical variations in audio inputs, and improving the system's ability to detect rare crop diseases. The ultimate goal is to contribute to the modernization of agricultural practices, offering farmers more reliable and effective tools to solve the challenges in crop management.

[Objective] The number, location, and crown spread of nursery stock are important foundations data for their scientific management. Traditional approach of conducting nursery stock inventories through on-site individual plant surveys is labor-intensive and time-consuming. Low-cost and convenient unmanned aerial vehicles (UAVs) for on-site collection of nursery stock data are beginning to be utilized, and the statistical analysis of nursery stock information through technical means such as image processing achieved. During the data collection process, as the flight altitude of the UAV increases, the number of trees in a single image also increases. Although the anchor box can cover more information about the trees, the cost of annotation is enormous in the case of a large number of densely populated tree images. To tackle the challenges of tree adhesion and scale variance in images captured by UAVs over nursery stock, and to reduce the annotation costs, using point-labeled data as supervisory signals, an improved dense detection and counting model was proposed to accurately obtain the location, size, and quantity of the targets. [Method] To enhance the diversity of nursery stock samples, the spruce dataset, the Yosemite, and the KCL-London publicly available tree datasets were selected to construct a dense nursery stock dataset. A total of 1 520 nursery stock images were acquired and divided into training and testing sets at a ratio of 7:3. To enhance the model's adaptability to tree data of different scales and variations in lighting, data augmentation methods such as adjusting the contrast and resizing the images were applied to the images in the training set. After enhancement, the training set consists of 3 192 images, and the testing set contains 456 images. Considering the large number of trees contained in each image, to reduce the cost of annotation, the method of selecting the center point of the trees was used for labeling. The LSC-CNN model was selected as the base model. This model can detect the quantity, location, and size of trees through point-supervised training, thereby obtaining more information about the trees. The LSC-CNN model was made improved to address issues of missed detections and false positives that occurred during the testing process. Firstly, to address the issue of missed detections caused by severe adhesion of densely packed trees, the last convolutional layer of the feature extraction network was replaced with dilated convolution. This change enlarges the receptive field of the convolutional kernel on the input while preserving the detailed features of the trees. So the model is better able to capture a broader range of contextual information, thereby enhancing the model's understanding of the overall scene. Secondly, the convolutional block attention module (CBAM) attention mechanism was introduced at the beginning of each scale branch. This allowed the model to focus on the key features of trees at different scales and spatial locations, thereby improving the model's sensitivity to multi-scale information. Finally, the model was trained using label smooth cross-entropy loss function and grid winner-takes-all strategy, emphasizing regions with highest losses to boost tree feature recognition. [Results and Discussions] The mean counting accuracy (MCA), mean absolute error (MAE), and root mean square error (RMSE) were adopted as evaluation metrics. Ablation studies and comparative experiments were designed to demonstrate the performance of the improved LSC-CNN model. The ablation experiment proved that the improved LSC-CNN model could effectively resolve the issues of missed detections and false positives in the LSC-CNN model, which were caused by the density and large-scale variations present in the nursery stock dataset. IntegrateNet, PSGCNet, CANet, CSRNet, CLTR and LSC-CNN models were chosen as comparative models. The improved LSC-CNN model achieved MCA, MAE, and RMSE of 91.23%, 14.24, and 22.22, respectively, got an increase in MCA by 6.67%, 2.33%, 6.81%, 5.31%, 2.09% and 2.34%, respectively; a reduction in MAE by 21.19, 11.54, 18.92, 13.28, 11.30 and 10.26, respectively; and a decrease in RMSE by 28.22, 28.63, 26.63, 14.18, 24.38 and 12.15, respectively, compared to the IntegrateNet, PSGCNet, CANet, CSRNet, CLTR and LSC-CNN models. These results indicate that the improved LSC-CNN model achieves high counting accuracy and exhibits strong generalization ability. [Conclusions] The improved LSC-CNN model integrated the advantages of point supervision learning from density estimation methods and the generation of target bounding boxes from detection methods.These improvements demonstrate the enhanced performance of the improved LSC-CNN model in terms of accuracy, precision, and reliability in detecting and counting trees. This study could hold practical reference value for the statistical work of other types of nursery stock.

[Objective] Nitrogen in soil is an absolutely crucial element for plant growth. Insufficient nitrogen supply can severely affect crop yield and quality, while excessive use of nitrogen fertilizers can lead to significant environmental issues such as water eutrophication and groundwater pollution. Therefore, large-scale, rapid detection of soil nitrogen content and precise fertilization are of great importance for smart agriculture. In this study, the hyperspectral data from the GF-5 satellite was emploied, and the various machine learning algorithms introduced to establish a prediction model for soil total nitrogen (TN) content and a distribution map of soil TN content was generated in the study area, aiming to provide scientific evidence for intelligent monitoring in smart agriculture. [Method] The study area was the Jian Sanjiang Reclamation Area in Fujin city, Heilongjiang province. Fieldwork involved the careful collection of 171 soil samples, obtaining soil spectral data, chemical analysis data of soil TN content, and the GF-5 hyperspectral data. Among these samples, 140 were randomly selected as the modeling sample set for calibration, and the remaining 31 samples were used as the test sample set. Three machine learning algorithms were introduced: Partial least squares regression (PLSR), backpropagation neural network (BPNN), and support vector machine (SVM) driven by a polynomial kernel function (Poly). Three distinct soil TN inversion models were constructed using these algorithms. To optimize model performance, ten-fold cross-validation was employed to determine the optimal parameters for each model. Additionally, multiple scatter correction (MSC) was applied to obtain band characteristic values, thus enhancing the model's prediction capability. Model performance was evaluated using three indicators: Coefficient of determination (R²), root mean square error (RMSE), and relative prediction deviation (RPD), to assess the prediction accuracy of different models. [Results and Discussions] MSC-Poly-SVM model exhibited the best prediction performance on the test sample set, with an R² of 0.863, an RMSE of 0.203, and an RPD of 2.147. This model was used to perform soil TN content inversion mapping using GF-5 satellite hyperspectral data. In accordance with the stringent requirements of land quality geochemical evaluation, the GF-5 hyperspectral land organic nitrogen parameter distribution map was drawn based on the "Determination of Land Quality Geochemical Evaluation". The results revealed that 86.1% of the land in the Jian Sanjiang study area had a total nitrogen content of more than 2.0 g/kg, primarily concentrated in first and second-grade plots, while third and fourth-grade plots accounted for only 11.83% of the total area. The study area exhibited sufficient soil nitrogen reserves, with high TN background values mainly concentrated along the riverbanks in the central part, distributed in a northeast-east direction. Specifically, in terms of soil spectral preprocessing, the median filtering method performed best in terms of smoothness and maintaining spectral characteristics. The spectra extracted from GF-5 imagery were generally quite similar to ground-measured spectral data, despite some noise, which had a minimal overall impact. [Conclusions] This study demonstrates the clear feasibility of using GF-5 satellite hyperspectral remote sensing data and machine learning algorithm for large-scale quantitative detection and visualization analysis of soil TN content. The soil TN content distribution map generated based on GF-5 hyperspectral remote sensing data is detailed and consistent with results from other methods, providing technical support for future large-scale quantitative detection of soil nutrient status and rational fertilization.

[Objective] Traditional object detection algorithms applied in the agricultural field, such as those used for crop growth monitoring and harvesting, often suffer from insufficient accuracy. This is particularly problematic for small crops like mushrooms, where recognition and detection are more challenging. The introduction of small object detection technology promises to address these issues, potentially enhancing the precision, efficiency, and economic benefits of agricultural production management. However, achieving high accuracy in small object detection has remained a significant challenge, especially when dealing with varying image sizes and target scales. Although the YOLO series models excel in speed and large object detection, they still have shortcomings in small object detection. To address the issue of maintaining high accuracy amid changes in image size and target scale, a novel detection model, Multi-Strategy Handling YOLOv8 (MSH-YOLOv8), was proposed. [Methods] The proposed MSH-YOLOv8 model builds upon YOLOv8 by incorporating several key enhancements aimed at improving sensitivity to small-scale targets and overall detection performance. Firstly, an additional detection head was added to increase the model's sensitivity to small objects. To address computational redundancy and improve feature extraction, the Swin Transformer detection structure was introduced into the input module of the head network, creating what was termed the "Swin Head (SH)". Moreover, the model integrated the C2f_Deformable convolutionv4 (C2f_DCNv4) structure, which included deformable convolutions, and the Swin Transformer encoder structure, termed "Swinstage", to reconstruct the YOLOv8 backbone network. This optimization enhanced feature propagation and extraction capabilities, increasing the network's ability to handle targets with significant scale variations. Additionally, the normalization-based attention module (NAM) was employed to improve performance without compromising detection speed or computational complexity. To further enhance training efficacy and convergence speed, the original loss function CIoU was replaced with wise-intersection over union (WIoU) Loss. Furthermore, experiments were conducted using mushrooms as the research subject on the open Fungi dataset. Approximately 200 images with resolution sizes around 600×800 were selected as the main research material, along with 50 images each with resolution sizes around 200×400 and 1 000×1 200 to ensure representativeness and generalization of image sizes. During the data augmentation phase, a generative adversarial network (GAN) was utilized for resolution reconstruction of low-resolution images, thereby preserving semantic quality as much as possible. In the post-processing phase, dynamic resolution training, multi-scale testing, soft non-maximum suppression (Soft-NMS), and weighted boxes fusion (WBF) were applied to enhance the model's small object detection capabilities under varying scales. [Results and Discussions] The improved MSH-YOLOv8 achieved an average precision at 50% (AP50) intersection over union of 98.49% and an AP@50-95 of 75.29%, with the small object detection metric APs reaching 39.73%. Compared to mainstream models like YOLOv8, these metrics showed improvements of 2.34%, 4.06% and 8.55%, respectively. When compared to the advanced TPH-YOLOv5 model, the improvements were 2.14%, 2.76% and 6.89%, respectively. The ensemble model, MSH-YOLOv8-ensemble, showed even more significant improvements, with AP50 and APs reaching 99.14% and 40.59%, respectively, an increase of 4.06% and 8.55% over YOLOv8. These results indicate the robustness and enhanced performance of the MSH-YOLOv8 model, particularly in detecting small objects under varying conditions. Further application of this methodology on the Alibaba Cloud Tianchi databases "Tomato Detection" and "Apple Detection" yielded MSH-YOLOv8-t and MSH-YOLOv8-a models (collectively referred to as MSH-YOLOv8). Visual comparison of detection results demonstrated that MSH-YOLOv8 significantly improved the recognition of dense and blurry small-scale tomatoes and apples. This indicated that the MSH-YOLOv8 method possesses strong cross-dataset generalization capability and effectively recognizes small-scale targets. In addition to quantitative improvements, qualitative assessments showed that the MSH-YOLOv8 model could handle complex scenarios involving occlusions, varying lighting conditions, and different growth stages of the crops. This demonstrates the practical applicability of the model in real-world agricultural settings, where such challenges are common. [Conclusions] The MSH-YOLOv8 improvement method proposed in this study effectively enhances the detection accuracy of small mushroom targets under varying image sizes and target scales. This approach leverages multiple strategies to optimize both the architecture and the training process, resulting in a robust model capable of high-precision small object detection. The methodology's application to other datasets, such as those for tomato and apple detection, further underscores its generalizability and potential for broader use in agricultural monitoring and management tasks.

[Objective] Crop line extraction is critical for improving the efficiency of autonomous agricultural machines in the field. However, traditional detection methods struggle to maintain high accuracy and efficiency under challenging conditions, such as strong light exposure and weed interference. The aims are to develop an effective crop line extraction method by combining YOLOv8-G, Affinity Propagation, and the Least Squares method to enhance detection accuracy and performance in complex field environments. [Methods] The proposed method employs machine vision techniques to address common field challenges. YOLOv8-G, an improved object detection algorithm that combines YOLOv8 and GhostNetV2 for lightweight, high-speed performance, was used to detect the central points of crops. These points were then clustered using the Affinity Propagation algorithm, followed by the application of the Least Squares method to extract the crop lines. Comparative tests were conducted to evaluate multiple backbone networks within the YOLOv8 framework, and ablation studies were performed to validate the enhancements made in YOLOv8-G. [Results and Discussions] The performance of the proposed method was compared with classical object detection and clustering algorithms. The YOLOv8-G algorithm achieved average precision (AP) values of 98.22%, 98.15%, and 97.32% for corn detection at 7, 14, and 21 days after emergence, respectively. Additionally, the crop line extraction accuracy across all stages was 96.52%. These results demonstrate the model's ability to maintain high detection accuracy despite challenging conditions in the field. [Conclusions] The proposed crop line extraction method effectively addresses field challenges such as lighting and weed interference, enabling rapid and accurate crop identification. This approach supports the automatic navigation of agricultural machinery, offering significant improvements in the precision and efficiency of field operations.

[Objective] In order to intelligently monitor the distribution of agricultural land cover types, high-spectral cameras are usually mounted on drones to collect high-spectral data, followed by classification of the high-spectral data to automatically draw crop distribution maps. Different crops have similar shapes, and the same crop has significant differences in different growth stages, so the network model for agricultural land cover classification requires a high degree of accuracy. However, network models with high classification accuracy are often complex and cannot be deployed on hardware systems. In view of this problem, a lightweight high-low frequency enhanced Reluformer network (ReluformerN) was proposed in this research. [Methods] Firstly, an adaptive octave convolution was proposed, which utilized the softmax function to automatically adjust the spectral dimensions of high-frequency features and low-frequency features, effectively alleviating the influence of manually setting the spectral dimensions and benefiting the subsequent extraction of spatial and spectral domain features of hyperspectral images. Secondly, a Reluformer was proposed to extract global features, taking advantage of the fact that low-frequency information could capture global features. Reluformer replaced the softmax function with a function of quadratic computational complexity, and through theoretical and graphical analysised, Relu function, LeakRelu function, and Gelu function were compared, it was found that the ReLU function and the softmax function both had non-negativity, which could be used for feature relevance analysis. Meanwhile, the ReLU function has a linearization feature, which is more suitable for self-relevance analysis. Therefore, the ReLU self-attention mechanism was proposed, which used the ReLU function to perform feature self-attention analysis. In order to extract deep global features, multi-scale feature fusion was used, and the ReLU self-attention mechanism was used as the core to construct the multi-head ReLU self-attention mechanism. Similar to the transformer architecture, the Reluformer structure was built by combining multi-head ReLU self-attention mechanism, feedforward layers, and normalization layers. With Reluformer as the core, the Reluformer network (ReluformerN) was proposed. This network considered frequency from the perspective of high-frequency information, taking into account the local features of image high-frequency information, and used deep separable convolution to design a lightweight network for fine-grained feature extraction of high-frequency information. It proposed Reluformer to extract global features for low-frequency information, which represented the global features of the image. ReluformerN was experimented on three public high-spectral data sets (Indian Pines, WHU-Hi-LongKou and Salinas) for crop variety fine classification, and was compared with five popular classification networks (2D-CNN, HybirdSN, ViT, CTN and LSGA-VIT). [Results and Discussion] ReluformerN performed best in overall accuracy (OA), average accuracy (AA), and other accuracy evaluation indicators. In the evaluation indicators of model parameters, model computation (FLOPs), and model complexity, ReluformerN had the smallest number of parameters and was less than 0.3 M, and the lowest computation. In the visualization comparison, the classification effect diagram of the model using ReluformerN had clearer image edges and more complete morphological structures, with fewer classification errors. The validity of the adaptive octave convolution was verified by comparing it with the traditional eightfold convolution. The classification accuracy of the adaptive octave convolution was 0.1% higher than that of the traditional octave convolution. When the artificial parameters were set to different values, the maximum and minimum classification accuracies of the traditional octave convolution were about 0.3% apart, while those of the adaptive octave convolution were only 0.05%. This showed that the adaptive octave convolution not only had the highest classification accuracy, but was also less sensitive to the artificial parameter setting, effectively overcoming the influence of the artificial parameter setting on the classification result. To validated the Reluformer module, it was compared with transformer, LeakRelufromer, and Linformer in terms of accuracy evaluation metrics such as OA and AA. The Reluformer achieved the highest classification accuracy and the lowest model parameter count among these models. This indicated that Reluformer not only effectively extracted global features but also reduced computational complexity. Finally, the effectiveness of the high-frequency and low-frequency branch networks was verified. The effectiveness of the high-frequency and low-frequency feature extraction branches was verified, and the characteristics of the feature distribution after high-frequency feature extraction, after high-low frequency feature extraction, and after the classifier were displayed using a 2D t-sne, compared with the original feature distribution. It was found that after high-frequency feature extraction, similar features were generally clustered together, but the spacing between different features was small, and there were also some features with overlapping situations. After low-frequency feature extraction, it was obvious that similar features were clustered more tightly. After high-low frequency feature fusion, and after the classifier, it was obvious that similar features were clustered, and different types of features were clearly separated, indicating that high-low frequency feature extraction enhanced the classification effect. [Conclusion] This network achieves a good balance between crop variety classification accuracy and model complexity, and is expected to be deployed on hardware systems with limited resources in the future to achieve real-time classification functions.

[Objective] The current crop management faces the challenges of difficulty in capturing personalized needs and the lack of flexibility in the decision-making process. To address the limitations of conventional precision agriculture systems, optimize key aspects of agricultural production, including crop yield, labor efficiency, and water and fertilizer use, while ensure sustainability and adaptability to diverse farming conditions, in this research, an intelligent decision-making method was presents for personalized vegetable crop water and fertilizer management using large language model (LLM) by integrating user-specific preferences into decision-making processes through natural language interactions. [Methods] The method employed artificial intelligence techniques, combining natural language processing (NLP) and reinforcement learning (RL). Initially, LLM engaged users through structured dialogues to identify their unique preferences related to crop production goals, such as maximizing yield, reducing resource consumption, or balancing multiple objectives. These preferences were then modeled as quantifiable parameters and incorporated into a multi-objective optimization framework. To realize this framework, proximal policy optimization (PPO) was applied within a reinforcement learning environment to develop dynamic water and fertilizer management strategies. Training was conducted in the gym-DSSAT simulation platform, a system designed for agricultural decision support. The RL model iteratively learned optimal strategies by interacting with the simulation environment, adjusting to diverse conditions and balancing conflicting objectives effectively. To refine the estimation of user preferences, the study introduced a two-phase process comprising prompt engineering to guide user responses and adversarial fine-tuning for enhanced accuracy. These refinements ensured that user inputs were reliably transformed into structured decision-making criteria. Customized reward functions were developed for RL training to address specific agricultural goals. The reward functions account for crop yield, resource efficiency, and labor optimization, aligning with the identified user priorities. Through iterative training and simulation, the system dynamically adapted its decision-making strategies to varying environmental and operational conditions. [Results and Discussions] The experimental evaluation highlighted the system's capability to effectively personalize crop management strategies. Using simulations, the method demonstrated significant improvements over traditional approaches. The LLM-based model accurately captured user-specific preferences through structured natural language interactions, achieving reliable preference modeling and integration into the decision-making process. The system's adaptability was evident in its ability to respond dynamically to changes in user priorities and environmental conditions. For example, in scenarios emphasizing resource conservation, water and fertilizer use were significantly reduced without compromising crop health. Conversely, when users prioritized yield, the system optimized irrigation and fertilization schedules to enhance productivity. These results showcased the method's flexibility and its potential to balance competing objectives in complex agricultural settings. Additionally, the integration of user preferences into RL-based strategy development enabled the generation of tailored management plans. These plans aligned with diverse user goals, including maximizing productivity, minimizing resource consumption, and achieving sustainable farming practices. The system's multi-objective optimization capabilities allowed it to navigate trade-offs effectively, providing actionable insights for decision-making. The experimental validation also demonstrated the robustness of the PPO algorithm in training the RL model. The system's strategies were refined iteratively, resulting in consistent performance improvements across various scenarios. By leveraging LLM to capture nuanced user preferences and combining them with RL for adaptive decision-making, the method bridges the gap between generic precision agriculture solutions and personalized farming needs. [Conclusions] This study established a novel framework for intelligent decision-making in agriculture, integrating LLM with reinforcement learning to address personalized crop management challenges. By accurately capturing user-specific preferences and dynamically adapting to environmental and operational variables, the method offers a transformative approach to optimizing agricultural productivity and sustainability. Future work will focus on expanding the system's applicability to a wider range of crops and environmental contexts, enhancing the interpretability of its decision-making processes, and facilitating integration with real-world agricultural systems. These advancements aim to further refine the precision and impact of intelligent agricultural decision-making systems, supporting sustainable and efficient farming practices globally.

[Objective] During the feeding process of fish populations in aquaculture, the video image characteristics of floating extruded feed on the water surface undergo continuous variations due to a myriad of environmental factors and fish behaviors. These variations pose significant challenges to the accurate detection of feed particles, which is crucial for effective feeding management. To address these challenges and enhance the detection of floating extruded feed particles on the water surface, ,thereby providing precise decision support for intelligent feeding in intensive aquaculture modes, the YOLOv11-AP2S model, an advanced detection model was proposed. [Methods] The YOLOv11-AP2S model enhanced the YOLOv11 algorithm by incorporating a series of improvements to its backbone network, neck, and head components. Specifically, an attention for fine-grained categorization (AFGC) mechanism was introduced after the 10th layer C2PSA of the backbone network. This mechanism aimed to boost the model's capability to capture fine-grained features, which were essential for accurately identifying feed particles in complex environments with low contrast and overlapping objects. Furthermore, the C3k2 module was replaced with the VoV-GSCSP module, which incorporated more sophisticated feature extraction and fusion mechanisms. This replacement further enhanced the network's ability to extract relevant features and improve detection accuracy. To improve the model's detection of small targets, a P2 layer was introduced. However, adding a P2 layer may increase computational complexity and resource consumption, so the overall performance and resource consumption of the model must be carefully balanced. To maintain the model's real-time performance while improving detection accuracy, a lightweight VoV-GSCSP module was utilized for feature fusion at the P2 layer. This approach enabled the YOLOv11-AP2S model to achieve high detection accuracy without sacrificing detection speed or model lightweights, making it suitable for real-time applications in aquaculture. [Results and Discussions] The ablation experimental results demonstrated the superiority of the YOLOv11-AP2S model over the original YOLOv11 network. Specifically, the YOLOv11-AP2S model achieved a precision ( P) and recall ( R) of 78.70%. The mean average precision (mAP50) at an intersection over union (IoU) threshold of 0.5 was as high as 80.00%, and the F₁-Score had also reached 79.00%. These metrics represented significant improvements of 6.7%, 9.0%, 9.4% (for precision, as previously mentioned), and 8.0%, respectively, over the original YOLOv11 network. These improvements showed the effectiveness of the YOLOv11-AP2S model in detecting floating extruded feed particles in complex environments. When compared to other YOLO models, the YOLOv11-AP2S model exhibits clear advantages in detecting floating extruded feed images on a self-made dataset. Notably, under the same number of iterations, the YOLOv11-AP2S model achieved higher mAP50 values and lower losses, demonstrating its superiority in detection performance. This indicated that the YOLOv11-AP2S model strikes a good balance between learning speed and network performance, enabling it to efficiently and accurately detect images of floating extruded feed on the water surface. Furthermore, the YOLOv11-AP2S model's ability to handle complex detection scenarios, such as overlapping and adhesion of feed particles and occlusion by bubbles, was noteworthy. These capabilities were crucial for accurate detection in practical aquaculture environments, where such challenges were common and can significantly impair the performance of traditional detection systems. The improvements in detection accuracy and efficiency made the YOLOv11-AP2S model a valuable tool for intelligent feeding systems in aquaculture, as it could provide more reliable and timely information on fish feeding behavior. Additionally, the introduction of the P2 layer and the use of the lightweight VoV-GSCSP module for feature fusion at this layer contributed to the model's overall performance. These enhancements enabled the model to maintain high detection accuracy while keeping computational costs and resource consumption within manageable limits. This was particularly important for real-time applications in aquaculture, where both accuracy and efficiency were critical for effective feeding management. [Conclusions] The successful application of the YOLOv11-AP2S model in detecting floating extruded feed particles demonstrates its potential to intelligent feeding systems in aquaculture. By providing accurate and timely information on fish feeding behavior, the model can help optimize feeding strategies, reduce feed waste, and improve the overall efficiency and profitability of aquaculture operations. Furthermore, the model's ability to handle complex detection scenarios and maintain high detection accuracy while keeping computational costs within manageable limits makes it a practical and valuable tool for real-time applications in aquaculture. Therefore, the YOLOv11-AP2S model holds promise for wide application in intelligent aquaculture management, contributing to the sustainability and growth of the aquaculture industry.

[Significance] The beef cattle industry plays a pivotal role in the development of China's agricultural economy and the enhancement of people's dietary structure. However, there exists a substantial disparity in feeding management practices and economic efficiency of beef cattle industry compared to developed countries. While the beef cattle industry in China is progressing towards intensive, modern, and large-scale development, it encounters challenges such as labor shortage and rising labor costs that seriously affect its healthy development. The determination of animal physiological indicators plays an important role in monitoring animal welfare and health status. Therefore, leveraging data collected from various sensors as well as technologies like machine learning, data mining, and modeling analysis enables automatic acquisition of meaningful information on beef cattle physiological indicators for intelligent management of beef cattle. In this paper, the intelligent monitoring technology of physiological indicators in beef cattle breeding process and its application value are systematically summarized, and the existing challenges and future prospects of intelligent beef cattle breeding process in China are prospected. [Progress] The methods of obtaining information on beef cattle physiological indicators include contact sensors worn on the body and non-contact sensors based on various image acquisitions. Monitoring the exercise behavior of beef cattle plays a crucial role in disease prevention, reproduction monitoring, and status assessment. The three-axis accelerometer sensor, which tracks the amount of time that beef cattle spend on lying, walking, and standing, is a widely used technique for tracking the movement behavior of beef cattle. Through machine vision analysis, individual recognition of beef cattle and identification of standing, lying down, and straddling movements can also be achieved, with the characteristics of non-contact, stress-free, low cost, and generating high data volume. Body temperature in beef cattle is associated with estrus, calving, and overall health. Sensors for monitoring body temperature include rumen temperature sensors and rectal temperature sensors, but there are issues with their inconvenience. Infrared temperature measurement technology can be utilized to detect beef cattle with abnormal temperatures by monitoring eye and ear root temperatures, although the accuracy of the results may be influenced by environmental temperature and monitoring distance, necessitating calibration. Heart rate and respiratory rate in beef cattle are linked to animal diseases, stress, and pest attacks. Monitoring heart rate can be accomplished through photoelectric volume pulse wave measurement and monitoring changes in arterial blood flow using infrared emitters and receivers. Respiratory rate monitoring can be achieved by identifying different nostril temperatures during inhalation and exhalation using thermal infrared imaging technology. The ruminating behavior of beef cattle is associated with health and feed nutrition. Currently, the primary tools used to detect rumination behavior are pressure sensors and three-axis accelerometer sensors positioned at various head positions. Rumen acidosis is a major disease in the rapid fattening process of beef cattle, however, due to limitations in battery life and electrode usage, real-time pH monitoring sensors placed in the rumen are still not widely utilized. Changes in animal physiology, growth, and health can result in alterations in specific components within body fluids. Therefore, monitoring body fluids or surrounding gases through biosensors can be employed to monitor the physiological status of beef cattle. By processing and analyzing the physiological information of beef cattle, indicators such as estrus, calving, feeding, drinking, health conditions, and stress levels can be monitored. This will contribute to the intelligent development of the beef cattle industry and enhance management efficiency. While there has been some progress made in developing technology for monitoring physiological indicators of beef cattle, there are still some challenges that need to be addressed. Contact sensors consume more energy which affects their lifespan. Various sensors are susceptible to environmental interference which affects measurement accuracy. Additionally, due to a wide variety of beef cattle breeds, it is difficult to establish a model database for monitoring physiological indicators under different feeding conditions, breeding stages, and breeds. Furthermore, the installation cost of various intelligent monitoring devices is relatively high, which also limits its utilization coverage. [Conclusion and Prospects] The application of intelligent monitoring technology for beef cattle physiological indicators is highly significance in enhancing the management level of beef cattle feeding. Intelligent monitoring systems and devices are utilized to acquire physiological behavior data, which are then analyzed using corresponding data models or classified through deep learning techniques to promptly monitor subtle changes in physiological indicators. This enables timely detection of sick, estrus, and calving cattle, facilitating prompt measures by production managers, reducing personnel workload, and improving efficiency. The future development of physiological indicators monitoring technologies in beef cattle primarily focuses on the following three aspects: (1) Enhancing the lifespan of contact sensors by reducing energy consumption, decreasing data transmission frequency, and improving battery life. (2) Integrating and analyzing various monitoring data from multiple perspectives to enhance the accuracy and utility value. (3) Strengthening research on non-contact, high-precision and automated analysis technologies to promote the precise and intelligent development within the beef cattle industry.

[Objective] In order to summarize exemplary cases of high-quality development in regional smart agriculture and contribute strategies for the sustainable advancement of the national smart agriculture cause, the spatiotemporal characteristics and key driving factors of smart farms in the Yangtze River Economic Belt were studied. [Methods] Based on data from 11 provinces (municipalities) spanning the years 2014 to 2023, a comprehensive analysis was conducted on the spatio-temporal differentiation characteristics of smart farms in the Yangtze River Economic Belt using methods such as kernel density analysis, spatial auto-correlation analysis, and standard deviation ellipse. Including the overall spatial clustering characteristics, high-value or low-value clustering phenomena, centroid characteristics, and dynamic change trends. Subsequently, the geographic detector was employed to identify the key factors driving the spatio-temporal differentiation of smart farms and to discern the interactions between different factors. The analysis was conducted across seven dimensions: special fiscal support, industry dependence, human capital, urbanization, agricultural mechanization, internet infrastructure, and technological innovation. [Results and Discussions] Firstly, in terms of temporal characteristics, the number of smart farms in the Yangtze River Economic Belt steadily increased over the past decade. The year 2016 marked a significant turning point, after which the growth rate of smart farms had accelerated noticeably. The development of the upper, middle, and lower reaches exhibited both commonalities and disparities. Specifically, the lower sub-regions got a higher overall development level of smart farms, with a fluctuating upward growth rate; the middle sub-regions were at a moderate level, showing a fluctuating upward growth rate and relatively even provincial distribution; the upper sub-regions got a low development level, with a stable and slow growth rate, and an unbalanced provincial distribution. Secondly, in terms of spatial distribution, smart farms in the Yangtze River Economic Belt exhibited a dispersed agglomeration pattern. The results of global auto-correlation indicated that smart farms in the Yangtze River Economic Belt tended to be randomly distributed. The results of local auto-correlation showed that the predominant patterns of agglomeration were H-L and L-H types, with the distribution across provinces being somewhat complex; H-H type agglomeration areas were mainly concentrated in Sichuan, Hubei, and Anhui; L-L type agglomeration areas were primarily in Yunnan and Guizhou. The standard deviation ellipse results revealed that the mean center of smart farms in the Yangtze River Economic Belt had shifted from Anqing city in Anhui province in 2014 to Jingzhou city in Hubei province in 2023, with the spatial distribution showing an overall trend of shifting southwestward and a slow expansion toward the northeast and south. Finally, in terms of key driving factors, technological innovation was the primary critical factor driving the formation of the spatio-temporal distribution pattern of smart farms in the Yangtze River Economic Belt, with a factor explanatory degree of 0.311 1. Moreover, after interacting with other indicators, it continued to play a crucial role in the spatio-temporal distribution of smart farms, which aligned with the practical logic of smart farm development. Urbanization and agricultural mechanization levels were the second and third largest key factors, with factor explanatory degrees of 0.292 2 and 0.251 4, respectively. The key driving factors for the spatio-temporal differentiation of smart farms in the upper, middle, and lower sub-regions exhibited both commonalities and differences. Specifically, the top two key factors driver identification in the upper region were technological innovation (0.841 9) and special fiscal support (0.782 3). In the middle region, they were technological innovation (0.619 0) and human capital (0.600 1), while in the lower region, they were urbanization (0.727 6) and technological innovation (0.425 4). The identification of key driving factors and the detection of their interactive effects further confirmed that the spatio-temporal distribution characteristics of smart farms in the Yangtze River Economic Belt were the result of the comprehensive action of multiple factors. [Conclusions] The development of smart farms in the Yangtze River Economic Belt is showing a positive momentum, with both the total number of smart farms and the number of sub-regions experiencing stable growth. The development speed and level of smart farms in the sub-regions exhibit a differentiated characteristic of "lower reaches > middle reaches > upper reaches". At the same time, the overall distribution of smart farms in the Yangtze River Economic Belt is relatively balanced, with the degree of sub-regional distribution balance being "middle reaches (Hubei province, Hunan province, Jiangxi province are balanced) > lower reaches (dominated by Anhui) > upper reaches (Sichuan stands out)". The coverage of smart farm site selection continues to expand, forming a "northeast-southwest" horizontal diffusion pattern. In addition, the spatio-temporal characteristics of smart farms in the Yangtze River Economic Belt are the result of the comprehensive action of multiple factors, with the explanatory power of factors ranked from high to low as follows: Technological innovation > urbanization > agricultural mechanization > human capital > internet infrastructure > industry dependence > special fiscal support. Moreover, the influence of each factor is further strengthened after interaction. Based on these conclusions, suggestions are proposed to promote the high-quality development of smart farms in the Yangtze River Economic Belt. This study not only provides a theoretical basis and reference for the construction of smart farms in the Yangtze River Economic Belt and other regions, but also helps to grasp the current status and future trends of smart farm development.

[Objective] Agricultural drought has a negative impact on the development of agricultural production and even poses a threat to food security. To reduce disaster losses and ensure stable crop yields, accurately predicting and classifying agricultural drought severity based on the standardized soil moisture index (SSMI) is of significant importance. [Methods] An agricultural drought prediction model, GCN-BiGRU-STMHSA was proposed, which integrated a graph convolutional network (GCN), a bidirectional gated recurrent unit (BiGRU), and a multi-head self-attention (MHSA) mechanism, based on remote sensing data. In terms of model design, the proposed method first employed GCN to fully capture the spatial correlations among different meteorological stations. By utilizing GCN, a spatial graph structure based on meteorological stations was constructed, enabling the extraction and modeling of spatial dependencies between stations. Additionally, a spatial multi-head self-attention mechanism (S-MHSA) was introduced to further enhance the model's ability to capture spatial features. For temporal modeling, BiGRU was utilized as the time-series feature extraction module. BiGRU considers both forward and backward dependencies in time-series data, enabling a more comprehensive understanding of the temporal dynamics of agricultural drought. Meanwhile, a temporal multi-head self-attention mechanism (T-MHSA) was incorporated to enhance the model's capability to learn long-term temporal dependencies and improve prediction stability across different time scales. Finally, the model employed a fully connected layer to perform regression prediction of the SSMI. Based on the classification criteria for agricultural drought severity levels, the predicted SSMI values were mapped to the corresponding drought severity categories, achieving precise agricultural drought classification. To validate the effectiveness of the proposed model, the global land data assimilation system (GLDAS_2.1) dataset and conducted modeling and experiments was utilized on five representative meteorological stations in the North China Plain (Xinyang, Gushi, Fuyang, Huoqiu, and Dingyuan). Additionally, the proposed model was compared with multiple deep learning models, including GRU, LSTM, and Transformer, to comprehensively evaluate its performance in agricultural drought prediction tasks. The experimental design covered different forecasting horizons to analyze the model's generalization capability in both short-term and long-term predictions, thereby providing a more reliable early warning system for agricultural drought. [Results and Discussions] Experimental results demonstrated that the proposed GCN-BiGRU-STMHSA model outperforms baseline models in both SSMI prediction and agricultural drought classification tasks. Specifically, across the five study stations, the model achieved significantly lower mean absolute error (MAE) and root mean squared error (RMSE), while attaining higher coefficient of determination ( R²), classification accuracy (ACC), and F₁-Score ( F₁). Notably, at the Gushi station, the model exhibited the best performance in predicting SSMI 10 days ahead, achieving an MAE of 0.053, a RMSE of 0.071, a R² of 0.880, an ACC of 0.925, and a F₁ of 0.924. Additionally, the model's generalization capability was investigated under different forecasting horizons (7, 14, 21, and 28 days). Results indicated that the model achieved the highest accuracy in short-term predictions (7 days). Although errors increase slightly as the prediction horizon extends, the model maintained high classification accuracy even for long-term predictions (up to 28 days). This highlighted the model's robustness and effectiveness in agricultural drought prediction over varying time scales. [Conclusions] The proposed model achieves superior accuracy and generalization capability in agricultural drought prediction and classification. By effectively integrating spatial graph modeling, temporal sequence feature extraction, and self-attention mechanisms, the model outperforms conventional deep learning approaches in both short-term and long-term forecasting tasks. Its strong performance provides accurate drought early warnings, assisting agricultural management authorities in formulating efficient water resource management strategies and optimizing irrigation plans. This contributes to safeguarding agricultural production and mitigating the potential adverse effects of agricultural drought.

[Objective] The picking of famous and high-quality tea is a crucial link in the tea industry. Identifying and locating the tender buds of famous and high-quality tea for picking is an important component of the modern tea picking robot. Traditional neural network methods suffer from issues such as large model size, long training times, and difficulties in dealing with complex scenes. In this study, based on the actual scenario of the Xiqing Tea Garden in Hunan Province, proposes a novel deep learning algorithm was proposed to solve the precise segmentation challenge of famous and high-quality tea picking points. [Methods] The primary technical innovation resided in the amalgamation of a lightweight network architecture, MobilenetV2, with an attention mechanism known as efficient channel attention network (ECANet), alongside optimization modules including atrous spatial pyramid pooling (ASPP). Initially, MobilenetV2 was employed as the feature extractor, substituting traditional convolution operations with depth wise separable convolutions. This led to a notable reduction in the model's parameter count and expedited the model training process. Subsequently, the innovative fusion of ECANet and ASPP modules constituted the ECA_ASPP module, with the intention of bolstering the model's capacity for fusing multi-scale features, especially pertinent to the intricate recognition of tea shoots. This fusion strategy facilitated the model's capability to capture more nuanced features of delicate shoots, thereby augmenting segmentation accuracy. The specific implementation steps entailed the feeding of image inputs through the improved network, whereupon MobilenetV2 was utilized to extract both shallow and deep features. Deep features were then fused via the ECA_ASPP module for the purpose of multi-scale feature integration, reinforcing the model's resilience to intricate backgrounds and variations in tea shoot morphology. Conversely, shallow features proceeded directly to the decoding stage, undergoing channel reduction processing before being integrated with upsampled deep features. This divide-and-conquer strategy effectively harnessed the benefits of features at differing levels of abstraction and, furthermore, heightened the model's recognition performance through meticulous feature fusion. Ultimately, through a sequence of convolutional operations and upsampling procedures, a prediction map congruent in resolution with the original image was generated, enabling the precise demarcation of tea shoot harvesting points. [Results and Discussions] The experimental outcomes indicated that the enhanced DeepLabV3+ model had achieved an average Intersection over Union (IoU) of 93.71% and an average pixel accuracy of 97.25% on the dataset of tea shoots. Compared to the original model based on Xception, there was a substantial decrease in the parameter count from 54.714 million to a mere 5.818 million, effectively accomplishing a significant lightweight redesign of the model. Further comparisons with other prevalent semantic segmentation networks revealed that the improved model exhibited remarkable advantages concerning pivotal metrics such as the number of parameters, training duration, and average IoU, highlighting its efficacy and precision in the domain of tea shoot recognition. This considerable decreased in parameter numbers not only facilitated a more resource-economical deployment but also led to abbreviated training periods, rendering the model highly suitable for real-time implementations amidst tea garden ecosystems. The elevated mean IoU and pixel accuracy attested to the model's capacity for precise demarcation and identification of tea shoots, even amidst intricate and varied datasets, demonstrating resilience and adaptability in pragmatic contexts. [Conclusions] This study effectively implements an efficient and accurate tea shoot recognition method through targeted model improvements and optimizations, furnishing crucial technical support for the practical application of intelligent tea picking robots. The introduction of lightweight DeepLabV3+ not only substantially enhances recognition speed and segmentation accuracy, but also mitigates hardware requirements, thereby promoting the practical application of intelligent picking technology in the tea industry.

[Objective] Acurately determining the suitable sowing date for winter wheat is of great significance for improving wheat yield and ensuring national food security. Traditional visual interpretation method is not only time-consuming and labor-intensive, but also covers a relatively small area. Remote sensing monitoring, belongs to post-event monitoring, exhibits a time lag. The aim of this research is to use the temperature threshold method and accumulated thermal time requirements for wheat leaves appearance method to analyze the suitable sowing date for winter wheat in county-level towns under the influence of long-term sequence of climate warming. [Methods] The research area were various townships in Qihe county, Shandong province. Based on European centre for medium-range weather forecasts (ECMWF) reanalysis data from 1997 to 2022, 16 meteorological data grid points in Qihe county were selected. Firstly, the bilinear interpolation method was used to interpolate the temperature data of grid points into the approximate center points of each township in Qihe county, and the daily average temperatures for each township were obtained. Then, temperature threshold method was used to determine the final dates of stable passage through 18, 16, 14 and 0 ℃. Key sowing date indicators such as suitable sowing temperature for different wheat varieties, growing degree days (GDD)≥0 ℃ from different sowing dates to before overwintering, and daily average temperature over the years were used for statistical analysis of the suitable sowing date for winter wheat. Secondly, the accumulated thermal time requirements for wheat leaves appearance method was used to calculate the appropriate date of GDD for strong seedlings before winter by moving forward from the stable date of dropping to 0 ℃. Accumulating the daily average temperatures above 0 ℃ to the date when the GDD above 0 ℃ was required for the formation of strong seedlings of wheat, a range of ±3 days around this calculated date was considered the theoretical suitable sowing date. Finally, combined with actual production practices, the appropriate sowing date of winter wheat in various townships of Qihe county was determined under the trend of climate warming. [Results and Discussions] The results showed that, from November 1997 to early December 2022, winter and annual average temperatures in Qihe county had all shown an upward trend, and there was indeed a clear trend of climate warming in various townships of Qihe county. Judging from the daily average temperature over the years, the temperature fluctuation range in November was the largest in a year, with a maximum standard deviation was 2.61 ℃. This suggested a higher likelihood of extreme weather conditions in November. Therefore, it was necessary to take corresponding measures to prevent and reduce disasters in advance to avoid affecting the growth and development of wheat. In extreme weather conditions, it was limited to determine the sowing date only by temperature or GDD. In cold winter years, it was too one-sided to consider only from the perspective of GDD. It was necessary to expand the range of GDD required for winter wheat before overwintering based on temperature changes to ensure the normal growth and development of winter wheat. The suitable sowing date for semi winter wheat obtained by temperature threshold method was from October 4th to October 16th, and the suitable sowing date for winter wheat was from September 27th to October 4th. Taking into account the GDD required for the formation of strong seedlings before winter, the suitable sowing date for winter wheat was from October 3rd to October 13th, and the suitable sowing date for semi winter wheat was from October 15th to October 24th, which was consisted with the suitable sowing date for winter wheat determined by the accumulated thermal time requirements for wheat leaves appearance method. Considering the winter wheat varieties planted in Qihe county, the optimal sowing date for winter wheat in Qihe county was from October 3rd to October 16th, and the optimal sowing date was from October 5th to October 13th. With the gradual warming of the climate, the suitable sowing date for wheat in various townships of Qihe county in 2022 was later than that in 2002. However, the sowing date for winter wheat was still influenced by factors such as soil moisture, topography, and seeding quality. The suitable sowing date for a specific year still needed to be adjusted to local conditions and flexibly sown based on the specific situation of that year. [Conclusions] The experimental results proved the feasibility of the temperature threshold method and accumulated thermal time requirements for wheat leaves appearance method in determining the suitable sowing date for winter wheat. The temperature trend can be used to identify cold or warm winters, and the sowing date can be adjusted in a timely manner to enhance wheat yield and reduce the impact of excessively high or low temperatures on winter wheat. The research results can not only provide decision-making reference for winter wheat yield assessment in Qihe county, but also provide an important theoretical basis for scientifically arrangement of agricultural production.

[Objective] The monitoring of livestock grazing in natural pastures is a key aspect of the transformation and upgrading of large-scale breeding farms. In order to meet the demand for large-scale farms to achieve accurate real-time detection of a large number of sheep, a high-precision and easy-to-deploy small-target detection model: CSD-YOLOv8s was proposed to realize the real-time detection of small-targeted individual sheep under the high-altitude view of the unmanned aerial vehicle (UAV). [Methods] Firstly, a UAV was used to acquire video data of sheep in natural grassland pastures with different backgrounds and lighting conditions, and together with some public datasets downloaded formed the original image data. The sheep detection dataset was generated through data cleaning and labeling. Secondly, in order to solve the difficult problem of sheep detection caused by dense flocks and mutual occlusion, the SPPFCSPC module was constructed with cross-stage local connection based on the you only look once (YOLO)v8 model, which combined the original features with the output features of the fast spatial pyramid pooling network, fully retained the feature information at different stages of the model, and effectively solved the problem of small targets and serious occlusion of the sheep, and improved the detection performance of the model for small sheep targets. In the Neck part of the model, the convolutional block attention module (CBAM) convolutional attention module was introduced to enhance the feature information capture based on both spatial and channel aspects, suppressing the background information spatially and focusing on the sheep target in the channel, enhancing the network's anti-jamming ability from both channel and spatial dimensions, and improving the model's detection performance of multi-scale sheep under complex backgrounds and different illumination conditions. Finally, in order to improve the real-time and deploy ability of the model, the standard convolution of the Neck network was changed to a lightweight convolutional C2f_DS module with a changeable kernel, which was able to adaptively select the corresponding convolutional kernel for feature extraction according to the input features, and solved the problem of input scale change in the process of sheep detection in a more flexible way, and at the same time, the number of parameters of the model was reduced and the speed of the model was improved. [Results and Discussions] The improved CSD-YOLOv8s model exhibited excellent performance in the sheep detection task. Compared with YOLO, Faster R-CNN and other classical network models, the improved CSD-YOLOv8s model had higher detection accuracy and frames per second (FPS) of 87 f/s in the flock detection task with comparable detection speed and model size. Compared with the YOLOv8s model, Precision was improved from 93.0% to 95.2%, mAP was improved from 91.2% to 93.1%, and it had strong robustness to sheep targets with different degree of occlusion and different scales, which effectively solved the serious problems of missed and misdetection of sheep in the grassland pasture UAV-on-ground sheep detection task due to the small sheep targets, large background noise, and high degree of densification. misdetection serious problems. Validated by the PASCAL VOC 2007 open dataset, the CSD-YOLOv8s model proposed in this study improved the detection accuracy of 20 different objects, including transportation vehicles, animals, etc., especially in sheep detection, the detection accuracy was improved by 9.7%. [Conclusions] This study establishes a sheep dataset based on drone images and proposes a model called CSD-YOLOv8s for detecting grazing sheep in natural grasslands. The model addresses the serious issues of missed detections and false alarms in sheep detection under complex backgrounds and lighting conditions, enabling more accurate detection of grazing livestock in drone images. It achieves precise detection of targets with varying degrees of clustering and occlusion and possesses good real-time performance. This model provides an effective detection method for detecting sheep herds from the perspective of drones in natural pastures and offers technical support for large-scale livestock detection in breeding farms, with wide-ranging potential applications.

[Objective] Grape picking is a key link in increasing production. However, in this process, a large amount of manpower and material resources are required, which makes the picking process complex and slow. To enhance harvesting efficiency and achieve automated grape harvesting, an improved YOLOv8n object detection model named 3C-YOLOv8n was proposed, which integrates the RealSense D415 depth camera for grape recognition and localization. [Methods] The propoesed 3C-YOLOv8n incorporated a convolutional block attention module (CBAM) between the first C2f module and the third Conv module in the backbone network. Additionally, a channel attention (CA) module was added at the end of the backbone structure, resulting in a new 2C-C2f backbone network architecture. This design enabled the model to sequentially infer attention maps across two independent dimensions (channel and spatial), optimize features by considering relationships between channels and positional information. The network structure was both flexible and lightweight. Furthermore, the Content-aware ReAssembly of Features up sampling operator was implemented to support instance-specific kernels (such as deconvolution) for feature reconstruction with neighboring pixels, replacing the nearest neighbor interpolation operator in the YOLOv8n neck network. This enhancement increased the receptive field and guided the reconstruction process based on input features while maintaining low parameter and computational complexity, thereby forming the 3C-YOLOv8n model. The pyrealsense2 library was utilized to obtain pixel position information from the target area using the Intel RealSense D415 camera. During this process, the depth camera was used to capture images, and target detection algorithms were employed to pinpoint the location of grapes. The camera's depth sensor facilitated the acquisition of the three-dimensional point cloud of grapes, allowing for the calculation of the distance from the pixel point to the camera and the subsequent determination of the three-dimensional coordinates of the center of the target's bounding box in the camera coordinate system, thus achieving grape recognition and localization. [Results and Discussions] Comparative and ablation experiments were conducted. it was observed that the 3C-YOLOv8n model achieved a mean average precision (mAP) of 94.3% at an intersection ratio of 0.5 (IOU=0.5), surpassing the YOLOv8n model by 1%. The accuracy (P) and recall (R) rates were recorded at 91.6% and 86.4%, respectively, reflecting increases of 0.1% and 0.7%. The F₁-Score also improved by 0.4%, demonstrating that the improved network model met the experimental accuracy and recall requirements. In terms of loss, the 3C-YOLOv8n algorithm exhibited superior performance, with a rapid decrease in loss values and minimal fluctuations, ultimately leading to a minimized loss value. This indicated that the improved algorithm quickly reached a convergence state, enhancing both model accuracy and convergence speed. The ablation experiments revealed that the original YOLOv8n model yielded a mAP of 93.3%. The integration of the CBAM and CA attention mechanisms into the YOLOv8n backbone resulted in mAP values of 93.5% each. The addition of the Content-aware ReAssembly of Features up sampling operator to the neck network of YOLOv8n produced a 0.5% increase in mAP, culminating in a value of 93.8%. The combination of the three improvement strategies yielded mAP increases of 0.3, 0.7, and 0.8%, respectively, compared to the YOLOv8n model. Overall, the 3C-YOLOv8n model demonstrated the best detection performance, achieving the highest mAP of 94.3%. The ablation results confirmed the positive impact of the proposed improvement strategies on the experimental outcomes. Compared to other mainstream YOLO series algorithms, all evaluation metrics showed enhancements, with the lowest missed detection and false detection rates among all tested algorithms, underscoring its practical advantages in detection tasks. [Conclusions] By effectively addressing the inefficiencies of manual labor, 3C-YOLOv8n network model not only enhances the precision of grape recognition and localization but also significantly optimizes overall harvesting efficiency. Its superior performance in evaluation metrics such as precision, recall, mAP, and F₁-Score, alongside the lowest recorded loss values among YOLO series algorithms, indicates a remarkable advancement in model convergence and operational effectiveness. Furthermore, the model's high accuracy in grape target recognition not only lays the groundwork for automated harvesting systems but also enables the implementation of complementary intelligent operations.

Objective Rice diseases significantly impact both the yield and quality of rice production. Automatic recognition of rice diseases using computer vision is crucial for ensuring high yields, quality, and efficiency. However, rice disease image recognition faces challenges such as limited availability of datasets, insufficient sample sizes, and imbalanced sample distributions across different disease categories. To address these challenges, a data augmentation method for rice leaf disease images was proposed based on an improved CycleGAN model in this reseach which aimed to expand disease image datasets by generating disease features, thereby alleviating the burden of collecting real disease data and providing more comprehensive and diverse data to support automatic rice disease recognition. Methods The proposed approach built upon the CycleGAN framework, with a key modification being the integration of a convolutional block attention module (CBAM) into the generator's residual module. This enhancement strengthened the network's ability to extract both local key features and global contextual information pertaining to rice disease-affected areas. The model increased its sensitivity to small-scale disease targets and subtle variations between healthy and diseased domains. This design effectively mitigated the potential loss of critical feature information during the image generation process, ensuring higher fidelity in the resulting images. Additionally, skip connections were introduced between the residual modules and the CBAM. These connections facilitate improved information flow between different layers of the network, addressing common issues such as gradient vanishing during the training of deep networks. Furthermore, a perception similarity loss function, designed to align with the human visual system, was incorporated into the overall loss function. This addition enabled the deep learning model to more accurately measure perceptual differences between the generated images and real images, thereby guiding the network towards producing higher-quality samples. This adjustment also helped to reduce visual artifacts and excessive smoothing, while concurrently improving the stability of the model during the training process. To comprehensively evaluate the quality of the rice disease images generated by the proposed model and to assess its impact on disease recognition performance, both subjective and objective evaluation metrics were utilized. These included user perception evaluation (UPE), structural similarity index (SSIM), peak signal-to-noise ratio (PSNR), and the performance of disease recognition within object detection frameworks. Comparative experiments were conducted across multiple GAN models, enabling a thorough assessment of the proposed model's performance in generating rice disease images. Additionally, different attention mechanisms, including efficient channel attention (ECA), coordinate attention (CA), and CBAM, were individually embedded into the generator's residual module. These variations allowed for a detailed comparison of the effects of different attention mechanisms on network performance and the visual quality of the generated images. Ablation studies were further performed to validate the effectiveness of the CBAM residual module and the perception similarity loss function in the network's overall architecture. Based on the generated rice disease samples, transfer learning experiments were conducted using various object detection models. By comparing the performance of these models before and after transfer learning, the effectiveness of the generated disease image data in enhancing the performance of object detection models was empirically verified. Results and Discussions The rice disease images generated by the improved CycleGAN model surpassed those produced by other GAN variants in terms of image detail clarity and the prominence of disease-specific features. In terms of objective quality metrics, the proposed model exhibited a 3.15% improvement in SSIM and an 8.19% enhancement in PSNR compared to the original CycleGAN model, underscoring its significant advantage in structural similarity and signal-to-noise ratio. The comparative experiments involving different attention mechanisms and ablation studies revealed that embedding the CBAM into the generator effectively increased the network's focus on critical disease-related features, resulting in more realistic and clearly defined disease-affected regions in the generated images. Furthermore, the introduction of the perception similarity loss function substantially enhanced the network's ability to perceive and represent disease-related information, thereby improving the visual fidelity and realism of the generated images. Additionally, transfer learning applied to object detection models such as YOLOv5s, YOLOv7-tiny, and YOLOv8s led to significant improvements in disease detection performance on the augmented dataset. Notably, the detection accuracy of the YOLOv5s model increased from 79.7% to 93.8%, representing a considerable enhancement in both generalization ability and robustness. This improvement also effectively reduced the rates of false positives and false negatives, resulting in more stable and reliable performance in rice disease detection tasks. Conclusions The rice leaf disease image generation method based on the improved CycleGAN model, as proposed in this study, effectively transforms images of healthy leaves into those depicting disease symptoms. By addressing the challenge of insufficient disease samples, this method significantly improves the disease recognition capabilities of object detection models. Therefore, it holds considerable application potential in the domain of leaf disease image augmentation and offers a promising new direction for expanding datasets of disease images for other crops.

[Objective] Accurate monitoring of rice planting areas is vital for ensuring national food security, evaluating greenhouse gas emissions, optimizing water resource allocation, and maintaining agricultural ecosystems. In recent years, the integration of remote sensing technologies—particularly the fusion of optical and synthetic aperture radar (SAR) data—has significantly enhanced the capacity to monitor crop distribution, even under challenging weather conditions. However, many current studies still rely heavily on phenological features captured at specific key stages, such as the transplanting phase, while overlooking the complete temporal dynamics of vegetation and water-related indices throughout the entire rice growth cycle. There is an urgent need for a method that fully leverages the time-series characteristics of remote sensing indices to enable accurate, scalable, and timely rice mapping. [Methods] Focusing on the Hangjiahu Plain, a typical rice-growing region in eastern China, a novel approach—dynamic NDVI-SDWI Fusion method for rice mapping (DNSF-Rice) was proposed in this research to accurately extract rice planting areas by synergistically integrating Sentinel-1 SAR and Sentinel-2 optical imagery on the google earth engine (GEE) platform. The methodological framework included the following three steps: First, using Sentinel-2 imagery, a time series of the normalized difference vegetation index (NDVI) was constructed. By analyzing its temporal dynamics across key rice growth stages, potential rice planting areas were identified through a threshold-based classification method; Second, a time series of the Sentinel-1 dual-polarized water index (SDWI) was generated to analyze its dynamic changes throughout the rice growth cycle. A thresholding algorithm was then applied to extract rice field distribution based on microwave data, considering the significant irrigation involved in rice cultivation; Finally, the spatial intersection of the NDVI-derived and SDWI-derived results was intersected to generate the final rice planting map. This step ensures that only pixels exhibiting both vegetation growth and irrigation signals were classified as rice. The classification datasets spanned five consecutive years from 2019 to 2023, with a spatial resolution of 10 m. [Results and Discussions] The proposed method demonstrated high accuracy and robust performance in mapping rice planting areas. Over the study period, the method achieved an overall accuracy of over 96% and an F₁-Score exceeding 0.96, outperforming several benchmark products in terms of spatial consistency and precision. The integration of NDVI and SDWI time-series features enabled effective identification of rice fields, even under the challenging conditions of frequent cloud cover and variable precipitation typical in the study area. Interannual analysis revealed a consistent increase in rice planting areas across the Hangjiahu Plain from 2019 to 2023. The remote sensing-based rice area estimates were in strong agreement with official agricultural statistics, further validating the reliability of the proposed method. The fusion of optical and SAR data proved to be a valuable strategy, effectively compensating for the limitations inherent in single-source imagery, especially during the cloudy and rainy seasons when optical imagery alone was often insufficient. Furthermore, the use of GEE facilitated the rapid processing of large-scale time-series data, supporting the operational scalability required for regional rice monitoring. This study emphasized the critical importance of capturing the full temporal dynamics of both vegetation and water signals throughout the entire rice growth cycle, rather than relying solely on fixed phenological stages. [Conclusions] By leveraging the complementary advantages of optical and SAR imagery and utilizing the complete time-series behavior of NDVI and SDWI indices, the proposed approach successfully mapped rice planting areas across a complex monsoon climate region over a five-year period. The method has been proven to be stable, reproducible, and adaptable for large-scale agricultural monitoring applications.

[Significance] Grass damage in farmland seriously restricts the quality and yield of crop planting and production, and promotes the occurrence of pests and diseases. Weed control is a necessary measure for high yield and high quality of crops. Currently, there are five main weed control methods: Manual, biological, thermal, mechanical, and chemical weed control. Traditional chemical weed control methods are gradually limited due to soil pollution and ecological balance disruption. Intelligent laser weeding technology, with the characteristics of environmental protection, high efficiency, flexibility, and automation, as an emerging and promising ecological and environmental protection new object control method for field weeds, has become the core direction to replace chemical weeding in recent years. The laser weeding robot is the carrier of laser weeding technology, an important manifestation of the development of modern agriculture towards intelligence and precision, and has great application and promotion value. [Progress] Laser weeding is currently a research hotspot to develop and study key technologies and equipment for smart agriculture, and has achieved a series of significant results, greatly promoting the promotion and application of intelligent laser weeding robots in the field. Laser weed control technology achieves precise weed control through thermal, photochemical, and photodynamic effects. In this article, the research background of laser weeding was introduced, its key technologies, operation system and equipment were discussed in details, covering aspects such as operating principles, system architecture, seedling, weed recognition and localization, robot navigation and path planning, as well as actuator control technologies. Then, based on the current research status of laser weeding robots, the existing problems and development trends of intelligent laser weeding robots were prospected. [Conclusion and Prospect] Based on the different field grass conditions in different regions, a large number of indoor and outdoor experiments on laser weed control should be carried out in the future to further verify the technical effectiveness and feasibility of laser field weed control, providing support for the research and application of laser weed control equipment technology. Despite facing challenges such as high costs and poor environmental adaptability, with the integration of technologies such as artificial intelligence and the Internet of Things, as well as policy support, laser weeding is expected to become an important support for sustainable agricultural development.

[Objective] Plug tray seedling cultivation is a contemporary method known for its high germination rates, uniform seedling growth, shortened transplant recovery period, diminished pest and disease incidence, and enhanced labor efficiency. Despite these advantages, challenges such as missing or underdeveloped seedlings can arise due to seedling quality and environmental factors. To ensure uniformity and consistency of the seedlings, sorting is frequently necessary, and the adoption of automated seedling sorting technology can significantly reduce labor costs. Nevertheless, the overgrowth of seedlings within the plugs can effect the accuracy of detection algorithms. A method for grading and locating strawberry seedlings based on a lightweight YOLOv8s model was presented in this research to effectively mitigate the interference caused by overgrown seedlings. [Methods] The YOLOv8s model was selected as the baseline for detecting different categories of seedlings in the strawberry plug tray cultivation process, namely weak seedlings, normal seedlings, and plug holes. To improve the detection efficiency and reduce the model's computational cost, the layer-adaptive magnitude-based pruning(LAMP) score-based channel pruning algorithm was applied to compress the base YOLOv8s model. The pruning procedure involved using the dependency graph to derive the group matrices, followed by normalizing the group importance scores using the LAMP Score, and ultimately pruning the channels according to these processed scores. This pruning strategy effectively reduced the number of model parameters and the overall size of the model, thereby significantly enhancing its inference speed while maintaining the capability to accurately detect both seedlings and plug holes. Furthermore, a two-stage seedling-hole matching algorithm was introduced based on the pruned YOLOv8s model. In the first stage, seedling and plug hole bounding boxes were matched according to their the degree of overlap (Dp), resulting in an initial set of high-quality matches. This step helped minimize the number of potential matching holes for seedlings exhibiting overgrowth. Subsequently, before the second stage of matching, the remaining unmatched seedlings were ranked according to their potential matching hole scores (S), with higher scores indicating fewer potential matching holes. The seedlings were then prioritized during the second round of matching based on these scores, thus ensuring an accurate pairing of each seedling with its corresponding plug hole, even in cases where adjacent seedling leaves encroached into neighboring plug holes. [Results and Discussions] The pruning process inevitably resulted in the loss of some parameters that were originally beneficial for feature representation and model generalization. This led to a noticeable decline in model performance. However, through meticulous fine-tuning, the model's feature expression capabilities were restored, compensating for the information loss caused by pruning. Experimental results demonstrated that the fine-tuned model not only maintained high detection accuracy but also achieved significant reductions in FLOPs (86.3%) and parameter count (95.4%). The final model size was only 1.2 MB. Compared to the original YOLOv8s model, the pruned version showed improvements in several key performance metrics: precision increased by 0.4%, recall by 1.2%, mAP by 1%, and the F₁-Score by 0.1%. The impact of the pruning rate on model performance was found to be non-linear. As the pruning rate increased, model performance dropped significantly after certain crucial channels were removed. However, further pruning led to a reallocation of the remaining channels' weights, which in some cases allowed the model to recover or even exceed its previous performance levels. Consequently, it was necessary to experiment extensively to identify the optimal pruning rate that balanced model accuracy and speed. The experiments indicated that when the pruning rate reached 85.7%, the mAP peaked at 96.4%. Beyond this point, performance began to decline, suggesting that this was the optimal pruning rate for achieving a balance between model efficiency and performance, resulting in a model size of 1.2 MB. To further validate the improved model's effectiveness, comparisons were conducted with different lightweight backbone networks, including MobileNetv3, ShuffleNetv2, EfficientViT, and FasterNet, while retaining the Neck and Head modules of the original YOLOv8s model. Results indicated that the modified model outperformed these alternatives, with mAP improvements of 1.3%, 1.8%, 1.5%, and 1.1%, respectively, and F₁-Score increases of 1.5%, 1.8%, 1.1%, and 1%. Moreover, the pruned model showed substantial advantages in terms of floating-point operations, model size, and parameter count compared to these other lightweight networks. To verify the effectiveness of the proposed two-stage seedling-hole matching algorithm, tests were conducted using a variety of complex images from the test set. Results indicated that the proposed method achieved precise grading and localization of strawberry seedlings even under challenging overgrowth conditions. Specifically, the correct matching rate for normal seedlings reached 96.6%, for missing seedlings 84.5%, and for weak seedlings 82.9%, with an average matching accuracy of 88%, meeting the practical requirements of the strawberry plug tray cultivation process. [Conclusions] The pruned YOLOv8s model successfully maintained high detection accuracy while reducing computational costs and improving inference speed. The proposed two-stage seedling-hole matching algorithm effectively minimized the interference caused by overgrown seedlings, accurately locating and classifying seedlings of various growth stages within the plug tray. The research provides a robust and reliable technical solution for automated strawberry seedling sorting in practical plug tray cultivation scenarios, offering valuable insights and technical support for optimizing the efficiency and precision of automated seedling grading systems.

[Objective] The detection of lameness in dairy cows is an important issue that needs to be solved urgently in the process of large-scale dairy farming. Timely detection and effective intervention can reduce the culling rate of young dairy cows, which has important practical significance for increasing the milk production of dairy cows and improving the economic benefits of pastures. Due to the low efficiency and low degree of automation of traditional manual detection and contact sensor detection, the mainstream cow lameness detection method is mainly based on computer vision. The detection perspective of existing computer vision-based cow lameness detection methods is mainly side view, but the side view perspective has limitations that are difficult to eliminate. In the actual detection process, there are problems such as cows blocking each other and difficulty in deployment. The cow lameness detection method from the top view will not be difficult to use on the farm due to occlusion problems. The aim is to solve the occlusion problem under the side view. [Methods] In order to fully explore the movement undulations of the trunk of the cow and the movement information in the time dimension during the walking process of the cow, a cow lameness detection method was proposed from a top view based on fused spatiotemporal flow features. By analyzing the height changes of the lame cow in the depth video stream during movement, a spatial stream feature image sequence was constructed. By analyzing the instantaneous speed of the lame cow's body moving forward and swaying left and right when walking, optical flow was used to capture the instantaneous speed of the cow's movement, and a time flow characteristic image sequence was constructed. The spatial flow and time flow features were combined to construct a fused spatiotemporal flow feature image sequence. Different from traditional image classification tasks, the image sequence of cows walking includes features in both time and space dimensions. There would be a certain distinction between lame cows and non-lame cows due to their related postures and walking speeds when walking, so using video information analysis was feasible to characterize lameness as a behavior. The video action classification network could effectively model the spatiotemporal information in the input image sequence and output the corresponding category in the predicted result. The attention module Convolutional Block Attention Module (CBAM) was used to improve the PP-TSMv2 video action classification network and build the Cow-TSM cow lameness detection model. The CBAM module could perform channel weighting on different modes of cows, while paying attention to the weights between pixels to improve the model's feature extraction capabilities. Finally, cow lameness experiments were conducted on different modalities, different attention mechanisms, different video action classification networks and comparison of existing methods. The data was used for cow lameness included a total of 180 video streams of cows walking. Each video was decomposed into 100‒400 frames. The ratio of the number of video segments of lame cows and normal cows was 1:1. For the feature extraction of cow lameness from the top view, RGB images had less extractable information, so this work mainly used depth video streams. [Results and Discussions] In this study, a total of 180 segments of cow image sequence data were acquired and processed, including 90 lame cows and 90 non-lame cows with a 1:1 ratio of video segments, and the prediction accuracy of automatic detection method for dairy cow lameness based on fusion of spatiotemporal stream features reaches 88.7%, the model size was 22 M, and the offline inference time was 0.046 s. The prediction accuracy of the common mainstream video action classification models TSM, PP-TSM, SlowFast and TimesFormer models on the data set of automatic detection method for dairy cow lameness based on fusion of spatiotemporal stream features reached 66.7%, 84.8%, 87.1% and 85.7%, respectively. The comprehensive performance of the improved Cow-TSM model in this paper was the most. At the same time, the recognition accuracy of the fused spatiotemporal flow feature image was improved by 12% and 4.1%, respectively, compared with the temporal mode and spatial mode, which proved the effectiveness of spatiotemporal flow fusion in this method. By conducting ablation experiments on different attention mechanisms of SE, SK, CA and CBAM, it was proved that the CBAM attention mechanism used has the best effect on the data of automatic detection method for dairy cow lameness based on fusion of spatiotemporal stream features. The channel attention in CBAM had a better effect on fused spatiotemporal flow data, and the spatial attention could also focus on the key spatial information in cow images. Finally, comparisons were made with existing lameness detection methods, including different methods from side view and top view. Compared with existing methods in the side-view perspective, the prediction accuracy of automatic detection method for dairy cow lameness based on fusion of spatiotemporal stream features was slightly lower, because the side-view perspective had more effective cow lameness characteristics. Compared with the method from the top view, a novel fused spatiotemporal flow feature detection method with better performance and practicability was proposed. [Conclusions] This method can avoid the occlusion problem of detecting lame cows from the side view, and at the same time improves the prediction accuracy of the detection method from the top view. It is of great significance for reducing the incidence of lameness in cows and improving the economic benefits of the pasture, and meets the needs of large-scale construction of the pasture.

[Objective] Sunlight-induced chlorophyll fluorescence (SIF) data obtained from satellites suffer from issues such as low spatial and temporal resolution, and discrete footprint because of the limitations imposed by satellite orbits. To address these problems, obtaining higher resolution SIF data, most reconstruction studies are based on low-resolution satellite SIF. Moreover, the spatial resolution of most SIF reconstruction products is still not enough to be directly used for the study of crop photosynthetic rate at the regional scale. Although some SIF products boast elevated resolutions, but these derive not from the original satellite SIF data reconstruct but instead evolve from secondary reconstructions based on preexisting SIF reconstruction products. Satellite OCO-2 (The Orbiting Carbon Obsevatory-2) equipped with a high-resolution spectrometer, OCO-2 SIF has higher spatial resolution (1.29×2.25 km) compared to other original SIF products, making it suitable in advancing the realm of high-resolution SIF data reconstruction, particularly within the context of regional-scale crop studies. [Methods] This research primarily exploration SIF reconstruct at the regional scale, mainly focused on the partial soybean planting regions nestled within the United States. The selection of MODIS raw data hinged on a meticulous consideration of environmental conditions, the distinctive physiological attributes of soybeans, and an exhaustive evaluation of factors intricately linked to OCO-2 SIF within these soybean planting regions. The primary tasks of this research encompassed reconstructing high resolution soybean SIF while concurrently executing a rigorous assessment of the reconstructed SIF's quality. During the dataset construction process, amalgamated SIF data from multiple soybean planting regions traversed by the OCO-2 satellite's footprint to retain as many of the available original SIF samples as possible. This approach provided the subsequent SIF reconstruction model with a rich source of SIF data. SIF data obtained beneath the satellite's trajectory were matched with various MODIS datasets, including enhanced vegetation index (EVI), fraction of photosynthetically active radiation (FPAR), and land surface temperature (LST), resulting in the creation of a multisource remote sensing dataset ultimately used for model training. Because of the multisource remote sensing dataset encompassed the most relevant explanatory variables within each SIF footprint coverage area concerning soybean physiological structure and environmental conditions. Through the activation functions in the BP neural network, it enhanced the understanding of the complex nonlinear relationships between the original SIF data and these MODIS products. Leveraging these inherent nonlinear relationships, compared and analyzed the effects of different combinations of explanatory variables on SIF reconstruction, mainly analyzing the three indicators of goodness of fit R², root mean square error RMSE, and mean absolute error MAE, and then selecting the best SIF reconstruction model, generate a regional scale, spatially continuous, and high temporal resolution (500 m, 8 d) soybean SIF reconstruction dataset (BPSIF). [Results and Discussions] The research findings confirmed the strong performance of the SIF reconstruction model in predicting soybean SIF. After simultaneously incorporating EVI, FPAR, and LST as explanatory variables to model, achieved a goodness of fit with an R² value of 0.84, this statistical metric validated the model's capability in predicting SIF data, it also reflected that the reconstructed 8 d time resolution of SIF data's reliability of applying to small-scale agricultural crop photosynthesis research with 500 m×500 m spatial scale. Based on this optimal model, generated the reconstructed SIF product (BPSIF). The Pearson correlation coefficient between the original OCO-2 SIF data and MODIS GPP stood were at a modest 0.53. In stark contrast, the correlation coefficient between BPSIF and MODIS Gross Primary Productivity (GPP) rosed significantly to 0.80. The increased correlation suggests that BPSIF could more accurately reflect the dynamic changes in GPP during the soybean growing season, making it more reliable compared to the original SIF data. Selected soybean planting areas in the United States with relatively single crop cultivation as the research area, based on high spatial resolution (1.29 km×2.25 km) OCO-2 SIF data, greatly reduced vegetation heterogeneity under a single SIF footprint. [Conclusions] The BPSIF proposed has significantly enhancing the regional and temporal continuity of OCO-2 SIF while preserving the time and spatial attributes contained in the original SIF dataset. Within the study area, BPSIF exhibits a significantly improved correlation with MODIS GPP compared to the original OCO-2 SIF. The proposed OCO-2 SIF data reconstruction method in this study holds the potential to provide a more reliable SIF dataset. This dataset has the potential to drive further understanding of soybean SIF at finer spatial and temporal scales, as well as find its relationship with soybean GPP.

[Objective] The body size parameter of cattle is a key indicator reflecting the physical development of cattle, and is also a key factor in the cattle selection and breeding process. In order to solve the demand of measuring body size of beef cattle in the complex environment of large-scale beef cattle ranch, an image acquisition device and an automatic measurement algorithm of body size were designed. [Methods] Firstly, the walking channel of the beef cattle was established, and when the beef cattle entered the restraining device through the channel, the RGB and depth maps of the image on the right side of the beef cattle were acquired using the Inter RealSense D455 camera. Secondly, in order to avoid the influence of the complex environmental background, an improved instance segmentation network based on Mask2former was proposed, adding CBAM module and CA module, respectively, to improve the model's ability to extract key features from different perspectives, extracting the foreground contour from the 2D image of the cattle, partitioning the contour, and comparing it with other segmentation algorithms, and using curvature calculation and other mathematical methods to find the required body size measurement points. Thirdly, in the processing of 3D data, in order to solve the problem that the pixel point to be measured in the 2D RGB image was null when it was projected to the corresponding pixel coordinates in the depth-valued image, resulting in the inability to calculate the 3D coordinates of the point, a series of processing was performed on the point cloud data, and a suitable point cloud filtering and point cloud segmentation algorithm was selected to effectively retain the point cloud data of the region of the cattle's body to be measured, and then the depth map was 16. Then the depth map was filled with nulls in the field to retain the integrity of the point cloud in the cattle body region, so that the required measurement points could be found and the 2D data could be returned. Finally, an extraction algorithm was designed to combine 2D and 3D data to project the extracted 2D pixel points into a 3D point cloud, and the camera parameters were used to calculate the world coordinates of the projected points, thus automatically calculating the body measurements of the beef cattle. [Results and Discussions] Firstly, in the part of instance segmentation, compared with the classical Mask R-CNN and the recent instance segmentation networks PointRend and Queryinst, the improved network could extract higher precision and smoother foreground images of cattles in terms of segmentation accuracy and segmentation effect, no matter it was for the case of occlusion or for the case of multiple cattles. Secondly, in three-dimensional data processing, the method proposed in the study could effectively extract the three-dimensional data of the target area. Thirdly, the measurement error of body size was analysed, among the four body size measurement parameters, the smallest average relative error was the height of the cross section, which was due to the more prominent position of the cross section, and the different standing positions of the cattle have less influence on the position of the cross section, and the largest average relative error was the pipe circumference, which was due to the influence of the greater overlap of the two front legs, and the higher requirements for the standing position. Finally, automatic body measurements were carried out on 137 beef cattle in the ranch, and the automatic measurements of the four body measurements parameters were compared with the manual measurements, and the results showed that the average relative errors of body height, cross section height, body slant length, and tube girth were 4.32%, 3.71%, 5.58% and 6.25%, respectively, which met the needs of the ranch. The shortcomings were that fewer body-size parameters were measured, and the error of measuring circumference-type body-size parameters was relatively large. Later studies could use a multi-view approach to increase the number of body rule parameters to be measured and improve the accuracy of the parameters in the circumference category. [Conclusions] The article designed an automatic measurement method based on two-dimensional and three-dimensional contactless body measurements of beef cattle. Moreover, the innovatively proposed method of measuring tube girth has higher accuracy and better implementation compared with the current research on body measurements in beef cattle. The relative average errors of the four body tape parameters meet the needs of pasture measurements and provide theoretical and practical guidance for the automatic measurement of body tape in beef cattle.

[Significance] Grain production spans multiple stages and involves numerous heterogeneous factors, including agronomic inputs, natural resources, environmental conditions, and socio-economic variables. However, the associated data generated throughout the entire production process, ranging from cultivation planning to harvest evaluation, remains highly fragmented, unstructured, and semantically diverse. This complexity data, combined with the lack of integrated core algorithms to support decision-making, has severely limited the potential of big data to drive innovation in grain production. Knowledge graph technology, by offering structured and semantically-rich representations of complex data, enables the integration of multi-source and heterogeneous data, enhances semantic mining and reasoning capabilities, and provides intelligent, knowledge-driven support for sustainable grain production, thereby addressing these challenges effectively. [Progress] This paper systematically reviewed the current research and application progress of knowledge graphs in the grain production big data. A comprehensive knowledge graph driven framework was proposed based on a hybrid paradigm combining data-driven modeling and domain knowledge guidance to support the entire grain production lifecycle and addressed three primary dimensions of data complexity: Structural diversity, relational heterogeneity, and semantic ambiguity. The key techniques of constructing multimodal knowledge map and temporal reasoning for grain production were described. First, an agricultural ontology system for grain production was designed, incorporating domain-specific concepts, hierarchical relationships, and attribute constraints. This ontology provided the semantic foundation for knowledge modeling and alignment. Second, multimodal named entity recognition (NER) techniques were employed to extract entities such as crops, varieties, weather conditions, operations, and equipment from structured and unstructured data sources, including satellite imagery, agronomic reports, Internet of Things sensor data, and historical statistics. Advanced deep learning models, such as bidirectional encoder representations from transformers (BERT) and vision-language transformers, were used to enhance recognition accuracy across text and image modalities. Third, the system implemented multimodal entity linking and disambiguation, which connected identical or semantically similar entities across different data sources by leveraging graph embeddings, semantic similarity measures, and rule-based matching. Finally, temporal reasoning modules were constructed using temporal knowledge graphs and logical rules to support dynamic inference over time-sensitive knowledge, such as crop growth stages, climate variations, and policy interventions. The proposed knowledge graph driven system enabled the development of intelligent applications across multiple stages of grain production. In the pre-production stage, knowledge graphs supported decision-making in resource allocation, crop variety selection, and planting schedule optimization based on past data patterns and predictive inference. During the in-production stage, the system facilitated precision operations, such as real-time fertilization and irrigation by reasoning over current field status, real-time sensor inputs, and historical trends. In the post-production stage, it enabled yield assessment and economic evaluation through integration of production outcomes, environmental factors, and policy constraints. [Conclusions and Prospects] Knowledge graph technologies offer a scalable and semantically-enhanced approach for unlocking the full potential of grain production big data. By integrating heterogeneous data sources, representing domain knowledge explicitly, and supporting intelligent reasoning, knowledge graphs can provide visualization, explainability, and decision support across various spatial scales, including national, provincial, county-level, and large-scale farm contexts. These technologies are of great scientific and practical significance in supporting China's national food security strategy and advancing the goals of storing grain in the land and storing grain in technology. Future directions include the construction of cross-domain agricultural knowledge fusion systems, dynamic ontology evolution mechanisms, and federated knowledge graph platforms for multi-region data collaboration under data privacy constraints.

[Objective] The rapid advancement of large language models (LLMs) has positioned them as a promising novel research paradigm in smart agriculture, leveraging their robust cognitive understanding and content generative capabilities. However, due to the lack of domain-specific agricultural knowledge, general LLMs often exhibit factual errors or incomplete information when addressing specialized queries, which is particularly prominent in agricultural applications. Therefore, enhancing the adaptability and response quality of LLMs in agricultural applications has become an important research direction. [Methods] To improve the adaptability and precision of LLMs in the agricultural applications, an innovative approach named the knowledge graph-guided agricultural LLM (KGLLM) was proposed. This method integrated information entropy for effective knowledge filtering and applied explicit constraints on content generation during the decoding phase by utilizing semantic information derived from an agricultural knowledge graph. The process began by identifying and linking key entities from input questions to the agricultural knowledge graph, which facilitated the formation of knowledge inference paths and the development of question-answering rationales. A critical aspect of this approach was ensuring the validity and reliability of the external knowledge incorporated into the model. This was achieved by evaluating the entropy difference in the model's outputs before and after the introduction of each piece of knowledge. Knowledge that didn't enhance the certainty of the answers was systematically filtered out. The knowledge paths that pass this entropy evaluation were used to adjust the token prediction probabilities, prioritizing outputs that were closely aligned with the structured knowledge. This allowed the knowledge graph to exert explicit guidance over the LLM's outputs, ensuring higher accuracy and relevance in agricultural applications. [Results and Discussions] The proposed knowledge graph-guided technique was implemented on five mainstream general-purpose LLMs, including open-source models such as Baichuan, ChatGLM, and Qwen. These models were compared with state-of-the-art knowledge graph-augmented generation methods to evaluate the effectiveness of the proposed approach. The results demonstrate that the proposed knowledge graph-guided approach significantly improved several key performance metrics of fluency, accuracy, factual correctness, and domain relevance. Compared to GPT-4o, the proposed method achieved notable improvements by an average of 2.592 3 in Mean BLEU, 2.815 1 in ROUGE, and 9.84% in BertScore. These improvements collectively signify that the proposed approach effectively leverages agricultural domain knowledge to refine the outputs of general-purpose LLMs, making them more suitable for agricultural applications. Ablation experiments further validated that the knowledge-guided agricultural LLM not only filtered out redundant knowledge but also effectively adjusts token prediction distributions during the decoding phase. This enhanced the adaptability of general-purpose LLMs in agriculture contexts and significantly improves the interpretability of their responses. The knowledge filtering and knowledge graph-guided model decoding method proposed in this study, which was based on information entropy, effectively identifies and selects knowledge that carried more informational content through the comparison of information entropy.Compared to existing technologies in the agricultural field, this method significantly reduced the likelihood of "hallucination" phenomena during the generation process. Furthermore, the guidance of the knowledge graph ensured that the model's generated responses were closely related to professional agricultural knowledge, thereby avoiding vague and inaccurate responses generated from general knowledge. For instance, in the application of pest and disease control, the model could accurately identify the types of crop diseases and corresponding control measures based on the guided knowledge path, thereby providing more reliable decision support. [Conclusions] This study provides a valuable reference for the construction of future agricultural large language models, indicating that the knowledge graphs guided mehtod has the potential to enhance the domain adaptability and answer quality of models. Future research can further explore the application of similar knowledge-guided strategies in other vertical fields to enhance the adaptability and practicality of LLMs across various professional domains.

[Objective] Winter wheat yield is crucial for national food security and the standard of living of the population. Existing crop yield prediction models often show low accuracy under disaster-prone climatic conditions. This study proposed an improved hierarchical linear model (IHLM) based on a drought weather index reduction rate, aiming to enhance the accuracy of crop yield estimation under drought conditions. [Methods] HLM was constructed using the maximum enhanced vegetation index-2 (EVI2max), meteorological data (precipitation, radiation, and temperature from March to May), and observed winter wheat yield data from 160 agricultural survey stations in Shandong province (2018－2021). To validate the model's accuracy, 70% of the data from Shandong province was randomly selected for model construction, and the remaining data was used to validate the accuracy of the yield model. HLM considered the variation in meteorological factors as a key obstacle affecting crop growth and improved the model by calculating the relative meteorological factors. The calculation of relative meteorological factors helped reduce the impact of inter-annual differences in meteorological data. The accuracy of the HLM model was compared with that of the random forest (RF), Support Vector Regression (SVR), and Extreme Gradient Boosting (XGBoost) models. The HLM model provided more intuitive interpretation, especially suitable for processing hierarchical data, which helped capture the variability of winter wheat yield data under drought conditions. Therefore, a drought weather index reduction rate model from the agricultural insurance industry was introduced to further optimize the HLM model, resulting in the construction of the IHLM model. The IHLM model was designed to improve crop yield prediction accuracy under drought conditions. Since the precipitation differences between Henan and Shandong provinces were small, to test the transferability of the IHLM model, Henan province sample data was processed in the same way as in Shandong, and the IHLM model was applied to Henan province to evaluate its performance under different geographical conditions. [Results and Discussions] The accuracy of the HLM model, improved based on relative meteorological factors (rMF), was higher than that of RF, SVR, and XGBoost. The validation accuracy showed a Pearson correlation coefficient (r) of 0.76, a root mean squared error (RMSE) of 0.60 t/hm², and a normalized RMSE (nRMSE) of 11.21%. In the drought conditions dataset, the model was further improved by incorporating the relationship between the winter wheat drought weather index and the reduction rate of winter wheat yield. After the improvement, the RMSE decreased by 0.48 t/hm², and the nRMSE decreased by 28.64 percentage points, significantly enhancing the accuracy of the IHLM model under drought conditions. The IHLM model also demonstrated good applicability when transferred to Henan province. [Conclusions] The IHLM model developed in this study improved the accuracy and stability of crop yield predictions, especially under drought conditions. Compared to RF, SVR, and XGBoost models, the IHLM model was more suitable for predicting winter wheat yield. This research can be widely applied in the agricultural insurance field, playing a significant role in the design of agricultural insurance products, rate setting, and risk management. It enables more accurate predictions of winter wheat yield under drought conditions, with results that are closer to actual outcomes.

[Significance] In view of the lack of monitoring means of quality influence factors in the production process of characteristic agricultural products with in central and western regions of China, the weak ability of intelligent control, the unclear coupling relationship of quality control elements and the low degree of systematic application, the existing technologies described such as intelligent monitoring of facility environment, growth and nutrition intelligent control model, architecture of intelligent management and control platform and so on. Through the application of the Internet of Things, big data and the new generation of artificial intelligence technology, it provides technical support for the construction and application of intelligent process quality control system for the whole growth period of characteristic agricultural products. [Progress] The methods of environmental regulation and nutrition regulation are analyzed, including single parameters and combined control methods, such as light, temperature, humidity, CO₂ concentration, fertilizer and water, etc. The multi-parameter coupling control method has the advantage of more comprehensive scene analysis. Based on the existing technology, a multi-factor coupling method of integrating growth state, agronomy, environment, input and agricultural work is put forward. This paper probes into the system architecture of the whole process service of quality control, the visual identification system of the growth process of agricultural products and the knowledge-driven agricultural technical service system, and introduces the technology of the team in the disease knowledge Q & A scene through multi-modal knowledge graph and large model technology. [Conclusions and Prospects] Based on the present situation of the production of characteristic facility agricultural products and the overall quality of farmers in the central and western regions of China, it is appropriate to transfer the whole technical system such as facility tomato, facility cucumber and so on. According to the varieties of characteristic agricultural products, cultivation models, quality control objectives to adapt to light, temperature, humidity and other parameters, as well as fertilizer, water, medicine and other input plans, a multi-factor coupling model suitable for a specific planting area is generated and long-term production verification and model correction are carried out. And popularize it in a wider area, making full use of the advantages of intelligent equipment and data elements will promote the realization of light simplification of production equipment, scene of intelligent technology, diversification of service models, on-line quality control, large-scale production of digital intelligence, and value of data elements, further cultivate facilities to produce new quality productivity.

[Objective] The enumeration of wheat leaves is an essential indicator for evaluating the vegetative state of wheat and predicting its yield potential. Currently, the process of wheat leaf counting in field settings is predominantly manual, characterized by being both time-consuming and labor-intensive. Despite advancements, the efficiency and accuracy of existing automated detection and counting methodologies have yet to satisfy the stringent demands of practical agricultural applications. This study aims to develop a method for the rapid quantification of wheat leaves to refine the precision of wheat leaf tip detection. [Methods] To enhance the accuracy of wheat leaf detection, firstly, an image dataset of wheat leaves across various developmental stages—seedling, tillering, and overwintering—under two distinct lighting conditions and using visible light images sourced from both mobile devices and field camera equipmen, was constructed. Considering the robust feature extraction and multi-scale feature fusion capabilities of YOLOv8 network, the foundational architecture of the proposed model was based on the YOLOv8, to which a coordinate attention mechanism has been integrated. To expedite the model's convergence, the loss functions were optimized. Furthermore, a dedicated small object detection layer was introduced to refine the recognition of wheat leaf tips, which were typically difficult for conventional models to discern due to their small size and resemblance to background elements. This deep learning network was named as YOLOv8-CSD, tailored for the recognition of small targets such as wheat leaf tips, ascertains the leaf count by detecting the number of leaf tips present within the image. A comparative analysis was conducted on the YOLOv8-CSD model in comparison with the original YOLOv8 and six other prominent network architectures, including Faster R-CNN, Mask R-CNN, YOLOv7, and SSD, within a uniform training framework, to evaluate the model's effectiveness. In parallel, the performance of both the original and YOLOv8-CSD models was assessed under challenging conditions, such as the presence of weeds, occlusions, and fluctuating lighting, to emulate complex real-world scenarios. Ultimately, the YOLOv8-CSD model was deployed for wheat leaf number detection in intricate field conditions to confirm its practical applicability and generalization potential. [Results and Discussions] The research presented a methodology that achieved a recognition precision of 91.6% and an mAP_0.5 of 85.1% for wheat leaf tips, indicative of its robust detection capabilities. This method exceled in adaptability within complex field environments, featuring an autonomous adjustment mechanism for different lighting conditions, which significantly enhanced the model's robustness. The minimal rate of missed detections in wheat seedlings' leaf counting underscored the method's suitability for wheat leaf tip recognition in intricate field scenarios, consequently elevating the precision of wheat leaf number detection. The sophisticated algorithm embedded within this model had demonstrated a heightened capacity to discern and focus on the unique features of wheat leaf tips during the detection process. This capability was essential for overcoming challenges such as small target sizes, similar background textures, and the intricacies of feature extraction. The model's consistent performance across diverse conditions, including scenarios with weeds, occlusions, and fluctuating lighting, further substantiated its robustness and its readiness for real-world application. [Conclusions] This research offers a valuable reference for accurately detecting wheat leaf numbers in intricate field conditions, as well as robust technical support for the comprehensive and high-quality assessment of wheat growth.

[Significance] The fluctuations in the supply, consumption, and prices of agricultural products directly affect market monitoring and early warning systems. With the ongoing transformation of China's agricultural production methods and market system, advancements in data acquisition technologies have led to an explosive growth in agricultural data. However, the complexity of the data, the narrow applicability of existing models, and their limited adaptability still present significant challenges in monitoring and forecasting the interlinked dynamics of multiple agricultural products. The efficient and accurate forecasting of agricultural market trends is critical for timely policy interventions and disaster management, particularly in a country with a rapidly changing agricultural landscape like China. Consequently, there is a pressing need to develop deep learning models that are tailored to the unique characteristics of Chinese agricultural data. These models should enhance the monitoring and early warning capabilities of agricultural markets, thus enabling precise decision-making and effective emergency responses. [Methods] An integrated forecasting methodology was proposed based on deep learning techniques, leveraging multi-dimensional agricultural data resources from China. The research introduced several models tailored to different aspects of agricultural market forecasting. For production prediction, a generative adversarial network and residual network collaborative model (GAN-ResNet) was employed. For consumption forecasting, a variational autoencoder and ridge regression (VAE-Ridge) model was used, while price prediction was handled by an Adaptive-Transformer model. A key feature of the study was the adoption of an "offline computing and visualization separation" strategy within the Chinese agricultural monitoring and early warning system (CAMES). This strategy ensures that model training and inference are performed offline, with the results transmitted to the front-end system for visualization using lightweight tools such as ECharts. This approach balances computational complexity with the need for real-time early warnings, allowing for more efficient resource allocation and faster response times. The corn, tomato, and live pig market data used in this study covered production, consumption and price data from 1980 to 2023, providing comprehensive data support for model training. [Results and Discussions] The deep learning models proposed in this study significantly enhanced the forecasting accuracy for various agricultural products. For instance, the GAN-ResNet model, when used to predict maize yield at the county level, achieved a mean absolute percentage error (MAPE) of 6.58%. The VAE-Ridge model, applied to pig consumption forecasting, achieved a MAPE of 6.28%, while the Adaptive-Transformer model, used for tomato price prediction, results in a MAPE of 2.25%. These results highlighted the effectiveness of deep learning models in handling complex, nonlinear relationships inherent in agricultural data. Additionally, the models demonstrate notable robustness and adaptability when confronted with challenges such as sparse data, seasonal market fluctuations, and heterogeneous data sources. The GAN-ResNet model excels in capturing the nonlinear fluctuations in production data, particularly in response to external factors such as climate conditions. Its capacity to integrate data from diverse sources—including weather data and historical yield data—made it highly effective for production forecasting, especially in regions with varying climatic conditions. The VAE-Ridge model addressed the issue of data sparsity, particularly in the context of consumption data, and provided valuable insights into the underlying relationships between market demand, macroeconomic factors, and seasonal fluctuations. Finally, the Adaptive-Transformer model stand out in price prediction, with its ability to capture both short-term price fluctuations and long-term price trends, even under extreme market conditions. [Conclusions] This study presents a comprehensive deep learning-based forecasting approach for agricultural market monitoring and early warning. The integration of multiple models for production, consumption, and price prediction provides a systematic, effective, and scalable tool for supporting agricultural decision-making. The proposed models demonstrate excellent performance in handling the nonlinearities and seasonal fluctuations characteristic of agricultural markets. Furthermore, the models' ability to process and integrate heterogeneous data sources enhances their predictive power and makes them highly suitable for application in real-world agricultural monitoring systems. Future research will focus on optimizing model parameters, enhancing model adaptability, and expanding the system to incorporate additional agricultural products and more complex market conditions. These improvements will help increase the stability and practical applicability of the system, thus further enhancing its potential for real-time market monitoring and early warning capabilities.

[Objective] Plant leaf shape is an important part of plant architectural model. Establishment of a three-dimensional structural model of leaves may assist simulating and analyzing plant growth. However, existing leaf modeling approaches lack interpretability, invertibility, and operability, which limit the estimation of model parameters, the simulation of leaf shape, the analysis and interpretation of leaf physiology and growth state, and model reusage. Aiming at the interoperability between three-dimensional structure representation and mathematical model parameters, this study paid attention to three aspects in wheat leaf shape parametric reconstruction: (1) parameter-driven model structure, (2) model parameter inversion, and (3) parameter dynamic mapping during growth. Based on this, a set of parameter-driven and point cloud inversion model for wheat leaf interoperability was proposed in this study. [Methods] A parametric surface model of a wheat leaf with seven characteristic parameters by using parametric modeling technology was built, and the forward parametric construction of the wheat leaf structure was realized. Three parameters, maximum leaf width, leaf length, and leaf shape factor, were used to describe the basic shape of the blade on the leaf plane. On this basis, two parameters, namely the angle between stems and leaves and the curvature degree, were introduced to describe the bending characteristics of the main vein of the blade in the three-dimensional space. Two parameters, namely the twist angle around the axis and the twist deviation angle around the axis, were introduced to represent the twisted structure of the leaf blade along the vein. The reverse parameter estimation module was built according to the surface model. The point cloud was divided by the uniform segmentation method along the Y-axis, and the veins were fit by a least squares regression method. Then, the point cloud was re-segmented according to the fit vein curve. Subsequently, the rotation angle was precisely determined through the segment-wise transform estimation method, with all parameters being optimally fit using the RANSAC regression algorithm. To validate the reliability of the proposed methodology, a set of sample parameters was randomly generated, from which corresponding sample point clouds were synthesized. These sample point clouds were then subjected to estimation using the described method. Then error analyzing was carried out on the estimation results. Three-dimensional imaging technology was used to collect the point clouds of Zhengmai 136, Yangmai 34, and Yanmai 1 samples. After noise reduction and coordinate registration, the model parameters were inverted and estimated, and the reconstructed point clouds were produced using the parametric model. The reconstruction error was validated by calculating the dissimilarity, represented by the Chamfer Distance, between the reconstructed point cloud and the measured point cloud. [Results and Discussions] The model could effectively reconstruct wheat leaves, and the average deviation of point cloud based parametric reconstruction results was about 1.2 mm, which had a high precision. Parametric modeling technology based on prior knowledge and point cloud fitting technology based on posterior data was integrated in this study to construct a digital twin model of specific species at the 3D structural level. Although some of the detailed characteristics of the leaves were moderately simplified, the geometric shape of the leaves could be highly restored with only a few parameters. This method was not only simple, direct and efficient, but also had more explicit geometric meaning of the obtained parameters, and was both editable and interpretable. In addition, the practice of using only tools such as rulers to measure individual characteristic parameters of plant organs in traditional research was abandoned in this study. High-precision point cloud acquisition technology was adopted to obtain three-dimensional data of wheat leaves, and pre-processing work such as point cloud registration, segmentation, and annotation was completed, laying a data foundation for subsequent research. [Conclusions] There is interoperability between the reconstructed model and the point cloud, and the parameters of the model can be flexibly adjusted to generate leaf clusters with similar shapes. The inversion parameters have high interpretability and can be used for consistent and continuous estimation of point cloud time series. This research is of great value to the simulation analysis and digital twinning of wheat leaves.