中国农业科学院农业信息研究所 中国农学会
Agricultural Information Institute of CAAS China Association of Agricultural Science Societies
En
Ch
中国农业科学院农业信息研究所 中国农学会 主办

AI检索

订阅

返回顶部
当前位置: 首页 > 文章 > 基于深度学习的松材线虫病害松木识别 林业工程学报 2021 (6) 142-147
Position: Home > Articles > Recognition of pine wood infected with pine nematode disease based on deep learning Journal of Forestry Engineering 2021 (6) 142-147

基于深度学习的松材线虫病害松木识别

作  者：
李浩;方伟泉;李浪浪;陈学永
单  位：
福建农林大学机电工程学院
关键词：
目标检测;病虫害;松材线虫病;无人机;深度学习
摘  要：
松材线虫病是世界上危害较大的森林病害之一,具有传染快、防治难的特点,严重威胁着我国的松木资源。在林区中识别、定位病害松木并及时进行治理是控制松材线虫病蔓延的有效手段。以小型商用无人机为平台获取林区遥感影像,分别对比了SSD、YOLO v3、Faster R-CNN 3种深度学习框架的训练效果,最终实现了遥感影像中病害松木的高效判别。考虑到病害松木在不同染病阶段存在病症差异性的情况,在标签数据集标记过程中将病害松木分为轻度、重度、病死三类。为提高模型的训练和识别效率,选择以效率更高的深度残差网络ResNet50代替VGG16作为深度学习框架的前置网络。试验结果表明,预训练模型调优技术的加入,有效减少了深度学习网络对前期数据量的依赖,对比SSD、YOLO v3目标检测框架,Faster R-CNN框架综合表现最好,对不同病害程度松木的识别正确率达到83.2%,实现了对林区遥感影像中病害松木高效精准的判别,也为林区病害松木的防治工作提供了可靠的辅助手段。
译  名：
Recognition of pine wood infected with pine nematode disease based on deep learning
作  者：
LI Hao;FANG Weiquan;LI Langlang;CHEN Xueyong;College of Mechanical and Electrical Engineering, Fujian Agriculture and Forestry University;
关键词：
object detection;;pests and diseases;;pine wood nematode disease;;UAV;;deep learning
摘  要：
The outbreak of pine wood nematode disease poses a huge threat to pine resources, which is characterized by rapid infection and difficult prevention and treatment, also known as “cancer of pine trees”. The outbreak of pine wood nematode disease would cause huge economic losses. In recent years, based on drone images, the emergence of image monitoring technology has gradually developed for addressing the time-consuming drawbacks of the traditional monitoring method of human surveys. Compared with the traditional machine learning technology, deep learning has gradually become an innovative technology of current research due to its strong technical advantages in the field of disease recognition. Although many classic algorithms have effectively improved the speed and accuracy of recognition, in the research field of pine wood nematode disease, there are relatively limited studies using deep learning frameworks. In this study, under the clear and windless weather, small commercial drones were used to obtain remote sensing image samples of forest areas while flying at the same altitude, and continuous sampling was performed at 25 d intervals. By comparing the three main target detection frameworks of Faster R-CNN, YOLO v3 and SSD, which are the most mainstream deep learning methods to identify the ward samples, it was found that the Faster R-CNN framework replaced VGG16 with ResNet50 as the front-end network had the best processing effect on diseased pines. Using the fine-tuning technology, the dependence on the amount of data in the early stage of deep learning network training was reduced effectively. The recognition accuracy rates of SSD and YOLO v3 frameworks for pine wood with different disease levels were 66.2% and 53.2%, respectively. Faster R-CNN had the best recognition result, with a recognition accuracy rate of 83.2%. This method not only overcame the misjudgments and omissions that were likely to occur in traditional identification methods, but also classified the disease levels according to mild, severe and death levels, achieving a more accurate identification of pine diseases in the disease control area and providing a reliable auxiliary method for forest pine prevention and control.

相似文章

导出引用
计量
文章访问数: 22
HTML全文浏览量: 0
PDF下载量: 0

所属期刊

推荐期刊

友情链接Friendly link
农业信息研究所Agricultural information institute 农业智能知识服务平台Agricultural intelligent knowledge service platform 农业学术服务平台Agricultural Knowledge Service System 中国知网CNKI 维普网VIP 万方WANFANG DATA
联系我们Contact us

电话Tell：010-82109652-808

邮箱E-mail：agrijournal@caas.cn

服务Service
关于我们About us 意见建议Opinions and suggest
版权说明Copyright
单位：中国农业科学院农业信息研究所 中国农学会 © 2017中国农业科学院农业信息研究所 ICP备案号：京ICP备09089781号-46