Classification of wheat stripe rust based on improved S-ResNet34 model
[Objectives]Rapid and accurate identification of wheat stripe rust disease grade is of great significance for its precise prevention and control.[Methods]In this paper,digital cameras were used to acquire RGB images of wheat stripe rust,and data sets of different disease levels of wheat stripe rust were constructed.Channel attention module(SE)and Inception module were added to the ResNet34 model to improve the model's attention and ability to extract the features of wheat stripe rust.And using the precision,recall,balance F score and accuracy and other evaluation indicators,the recognition accuracy of S-ResNet34 was compared with that of VGG16,MobileNetV2,Swin-Transformer,ResNet34 and other mainstream models.[Results]The training accuracy of S-ResNet34 model was 93.85％,compared with deep network models such as VGG16(84.53％),MobileNet V2(79.35％),Swin-Transformer(85.67％)and ResNet34(87.50％).The accuracy rate increased by 9.32％,14.50％,8.18％and 6.35％,respectively.At the same time,the model loss value was smaller,and the improved ResNet34 model had better ability to identify the damage characteristics of wheat stripe rust and faster training convergence.[Conclusions]The deep learning model can accurately identify the degree of wheat stripe rust disease,and adding attention module to ResNet34 model can effectively improve the identification accuracy of wheat stripe rust disease.
wheat stripe rustdeep learningdisease gradeimage recognitionimproved S-ResNet34 model
万方数据
维普
