基于深度残差的多特征多粒度农业病虫害识别研究  

Research on identification of agricultural insects based on depth residual network with multi-feature and multi-granularity

作　　者：李艳红[1] 樊同科[1] LI Yan-hong;FAN Tong-ke(Xi’an International University,Xi’an 710077,China)

机构地区：[1]西安外事学院,西安710077

出　　处：《湖北农业科学》2020年第16期153-157,共5页Hubei Agricultural Sciences

基　　金：陕西省2019年重点研发计划项目(2019NY-055);陕西省教育科学十三五规划课题(SGH18H538)。

摘　　要：为了实现复杂农田背景下的病虫害识别,提出了一种基于深度残差学习的多特征多粒度农业病虫害识别方法。结果表明,与传统SVM和BP神经网络相比,该算法在复杂农田背景下的病虫害图像识别精度明显提高。在复杂农田背景下10种作物病虫害图像的分类问题上取得了98.67%的精度。该算法具有很高的实际应用价值,可以与当前使用的农业联网系统集成到实际的农业病虫害防治中。In order to realize insect identification under complex farmland background,this paper proposed a multi-feature and multigranularity insect identification method based on deep residual network.Compared with the traditional SVM and BP neural network,the accuracy of insect identification based on deep residual network was significantly improved in complex farmland background.Com⁃pared with the deep convolutional neural network such as AlexNet,the performance of our method was further improved after the depth residual learning optimization.And the accuracy of 98.67%was obtained on the classification of crop insect images under the back⁃ground of 10 types of complex farmland.Therefore,this method has high practical application value and can be integrated into the actu⁃al agricultural insect control task with the currently used agricultural networking system.

关 键 词：深度残差 多特征多粒度 农业病虫害 识别 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] S435[自动化与计算机技术—计算机科学与技术]