当前位置: 首页 > 文章 > 基于卷积神经网络的树种识别研究 森林工程 2020 (1) 33-38
Position: Home > Articles > Study on Tree Species Identification Based on Convolution Neural Network Forest Engineering 2020 (1) 33-38

基于卷积神经网络的树种识别研究

作  者:
刘忠伟;戚大伟
单  位:
东北林业大学理学院
关键词:
卷积神经网络;树皮纹理图像;树种识别
摘  要:
由于森林资源的重要性和不可替代性,准确识别树种是研究和保护森林资源的基础。本研究采用ROI(感兴趣区域)截取及直方图均衡化的图像增强方法对原始数据集进行预处理,基于调整和优化的Lenet 5卷积神经网络模型结构,对无干扰背景下的水曲柳、家榆和白桦等5种典型东北林木的树皮纹理RGB图像自动提取特征,进行分类识别。结果表明,该卷积神经网络对5种树种的识别正确率达到95.8%。为林业资源管理节约人工定义树皮纹理特征的成本,为计算机自动识别树种提供更高效、更准确和鲁棒性更强的方法。
译  名:
Study on Tree Species Identification Based on Convolution Neural Network
作  者:
LIU Zhongwei;QI Dawei;College of Science, Northeast Forestry University;
关键词:
Convolution neural network;;bark texture images;;tree species identification
摘  要:
The study is conducted based on the importance and irreplaceability of forest resources, identifying tree species accurately is the basic to research and protect forest resources. In this study, ROI(Region of Interest) interception and histogram equalization is employed to preprocess the original image data. Based on the adjusted and optimized Lenet5 convolution neural network model structure, the network automatically extracts the RGB images of bark texture of five typical northeastern forests, including Fraxinus mandshurica, Ulmus pumila and white birch, in the background without disturbance, and then classifies and recognizes them. The results show that the recognition accuracy of the convolution neural network is 95.8%. This study saves the cost of manual definition of bark texture features for forestry resource management. It can provide more efficient, accurate and robust method for computer automatic identification of tree species.

相似文章

计量
文章访问数: 11
HTML全文浏览量: 0
PDF下载量: 0

所属期刊

推荐期刊