基于改进VGG网络的农作物病害图像识别  

Crop Disease Image Recognition Based on Improved VGG Network

作　　者：岳有军[1] 李雪松 赵辉[1,2] 王红君 Yue Youjun;Li Xuesong;Zhao Hui;Wang Hongjun(Tianjin University of Technology, Tianjin Key Laboratory of Complex System Control Theory and Application/School of Electrical and Electronic Engineering, Tianjin 300384, China;School of engineering and Technology, Tianjin Agricultural University, Tianjin 300392,China)

机构地区：[1]天津理工大学天津市复杂系统控制理论与应用重点实验室/电气电子工程学院,天津300384 [2]天津农学院工程技术学院,天津300392

出　　处：《农机化研究》2022年第6期18-24,共7页Journal of Agricultural Mechanization Research

基　　金：天津市科技计划项目(18YFZCNC01120)。

摘　　要：随着计算机技术的飞速发展,使用机器视觉进行农作物病害识别成为了一种趋势。但是,当前农作物病害图像识别研究主要集中在提高其识别精度方面而很少考虑实际复杂自然条件下的鲁棒性研究。在实际复杂自然条件下,噪声和复杂自然条件背景会降低识别精度。为此,对VGG网络进行改进,将高阶残差和参数共享反馈子网络添加进VGG网络中,识别实际复杂自然条件下的农作物病害。农作物病害表观的特征表达由高阶残差子网络提供,高阶残差子网络使病害识别的准确率更高;病害图像深层特征中的背景噪声被参数共享反馈子网络削弱,使改进VGG网络具有更强的鲁棒性。实验分析表明:在实际大田环境中,此方法在识别精度和鲁棒性方面比SVM、AlexNET、ResNet-50、VGG-16效果更好。With the rapid development of computer technology,using machine vision to identify crop diseases has become a trend.However,the current research on crop disease image recognition mainly focuses on improving the recognition accuracy,and seldom considers the robustness research under the actual complex natural conditions.In the actual complex natural conditions,noise and background of complex natural conditions will reduce the recognition accuracy.Therefore,in this paper,VGG network is improved by adding high-order residual and parameter sharing feedback sub network into VGG network to identify crop diseases under actual complex natural conditions.The feature expression of crop disease appearance is provided by high-order residual sub network,which makes the accuracy rate of disease recognition higher.The background noise in deep feature of disease image is weakened by parameter sharing feedback sub network,which makes the improved VGG network have stronger robustness.The experimental results show that the proposed method is better than SVM,alexnet,resnet-50 and vgg-16 in recognition accuracy and robustness.

关 键 词：农作物病害识别 VGG网络 高阶残差子网络 参数共享反馈子网络 

分 类 号：S126[农业科学—农业基础科学] S431