基于改进空间残差收缩网络模型的农作物病虫害识别  

The Recognition for Crop Pests and Diseases Based on the Improved Residual Shrinkage Network

作　　者：刘晓锋[1] 高丽梅 LIU Xiao-feng;GAO Li-mei(School of Automotive and Transportation/Tianjin University of Technology and Education,Tianjin 300222,China;Tianjin Academy of Transportation Science,Tianjin 300074,China)

机构地区：[1]天津职业技术师范大学汽车与交通学院,天津300222 [2]天津市交通科学研究院,天津300074

出　　处：《山东农业大学学报（自然科学版）》2022年第2期259-264,共6页Journal of Shandong Agricultural University：Natural Science Edition

摘　　要：为了提高农作物病虫害识别的精度,本文将3D-CNN和2D-CNN与空间残差网络相结合,软阈值化作为非线性层嵌入空间残差网络以消除病虫害图像不重要的图像特征,提出一种基于空间残差收缩网络的农作物病虫害识别模型。与3D-CNN和ResNet相比,基于空间残差收缩网络的农作物病虫害识别模型具有更高的精度和鲁棒性,总体分类精度为99.41%,增强了图像特征与病虫害类别的关系,可以识别多种农作物病虫害图像。In order to improve the precision of crop pests and diseases recognition,3D-CNN and 2D-CNN are combined with spatial residual shrinkage network(SRSN).As a non-linear layer embedded in SRSN,soft thresholding is used to eliminate the unimportant image features of crop pests and diseases.Compared with 3D-CNN and ResNet,the proposed SRSN model has higher accuracy and robustness,and the overall recognition accuracy is 99.41%.Moreover,the proposed model enhances the relationship between image features and crop pests and diseases recognition,which can be used to recognize different crop pests and diseases for images.

关 键 词：空间残差收缩网络 农作物病虫害 图像识别 

分 类 号：S126[农业科学—农业基础科学]