Research on cotton pest and disease identification method based on RegNet-CMTL

Cotton is an important crop in China, and it is important to realize accurate identification and classification of cotton pests and diseases to improve the quality of cotton. In this paper, a RegNet-CMTL algorithm is proposed, which not only effectively improves the accuracy of cotton pest and disease identification, but also significantly reduces the false alarm rate. First, the risk of overfitting is reduced by regularizing the data enhancement method. Meanwhile, CBAM attention mechanism and multi-scale feature fusion strategy are added to enhance the feature extraction and fine-grained classification ability of the model. Finally, a migration learning strategy is applied to further optimize the overall performance and convergence speed of the model. The experimental results show that the RegNet-CMTL model improves the average recognition accuracy by 5.416% and reduces the average false alarm rate by 1.110% compared to RegNet, and compares with other convolutional neural networks such as Shufflenet, ResNet34, VGG-16, Alexnet, and Googlenet, the average recognition accuracy is improved by 25.226%, 11.064%, 3.924% and 1.742%. Therefore, the RegNet-CMTL algorithm can effectively reduce the error of mixing pest and healthy cotton during the cotton picking process, thus improving the overall quality of cotton.