基于卷积神经网络的银杏叶片患病程度识别  

Identification of Leaf Disease Grade Based on Convolutional Neural Network

作　　者：刘瑾蓉 林剑辉[1] 李婷婷[1] LIU Jinrong;LIN Jianhui;LI Tingting(School of Technology, Beijing Forestry University, Beijing 100089, China)

机构地区：[1]北京林业大学工学院,北京100089

出　　处：《中国农业科技导报》2018年第6期55-61,共7页Journal of Agricultural Science and Technology

基　　金：北京林业大学中央高校基本科研业务费项目(2015ZCQ-GX-03);北京林业大学校级研究生专业实践基地建设项目资助;国家自然科学基金项目(31371537)

摘　　要：银杏是我国一种常见的经济林木,对银杏叶病害进行数字化辨识有助于提高银杏种植产业的管理水平,为其病害预警提供可能。以银杏轮纹病为研究对象,采用卷积神经网络为分类算法,对银杏患病程度进行自动辨识。根据银杏轮纹病的叶片特点,设计了18层卷积神经网络,核心功能主要由四个卷积层,四个池化层、两个全连接层提供。经过多次训练与测试,网络对银杏叶片的5种不同患病程度辨识率最低可达92%以上。将设计的卷积神经网络与传统的BP神经网络、Alex-Net网络进行对比实验,结果表明,卷积神经网络在银杏轮纹病患病程度辨识的应用上具有更高的精度。上述结果对于银杏其他病害或其他植物病害辨识应用中具有一定的借鉴意义。Ginkgo biloba is a common economic tree in China. Digital identification of Ginkgo biloba diseases can improve the management level of ginkgo plantation industry,and provide possibility of early warning for diseases.Taking Ginkgo biloba ring disease as research objective,this paper carried out automatic identity on the degree of Ginkgo biloba disease by convolutional neural network as a classification algorithm. According to the leaf characteristics of Gingko biloba ring disease,18-layer convolutional neural network was designed,including 4 feature extraction layers,4 max-pooling layers and 2 fully connected layers. The experimental results showed that the identification rate of 5 different disease degrees is above 92%. The convolutional neural network designed in this paper was compared with the traditional BP neural network and Alex-Net,and the results showed that convolutional neural network possessed higher accuracy in identifying Ginkgo biloba disease degrees. This conclusion was of certain referential significance for identifying other diseases of Ginkgo biloba or other plant disease.

关 键 词：卷积神经网络 银杏轮纹病 病患程度识别 BP神经网络 

分 类 号：S126[农业科学—农业基础科学]