利用改进MobileNet V2网络识别水稻叶片病虫害的方法

Rice leaf disease and insect pest recognition method based on improved MobileNet V2 network

针对传统水稻叶片病虫害分类效率不高、精度低和模型占用空间过大等问题,改进MobileNet V2网络,并结合迁移学习策略对水稻叶片病虫害进行识别。实验以4种常见水稻叶片病虫害作为研究对象,利用注意力机制对MobileNet V2进行改进,通过修改模型残差结构引入通道注意力机制,并采用迁移学习策略对改进模型进行训练。实验表明,相比于原始模型,引入注意力机制并采用迁移学习的改进模型CAM_qianyi的准确率提升了0.82个百分点,达到了84.32%,其准确率也高于轻量化卷积神经网络ResNet18(82.54%)和未采用迁移学习的改进模型CAM(73.65%)。改进模型能准确提取水稻叶片病虫害特征,有效提高了识别效率和精度。

In response to the issues of low classification efficiency,poor accuracy,and excessive model space occupation in traditional rice leaf disease and insect pest classification,this article improves the MobileNet V2 network and combining transfer learning strategies to identify rice leaf pests and diseases.In the experiment,four common rice leaf diseases and insect pests are taken as the research objects.The attention mechanism is used to improve MobileNet V2.By modifying the residual structure of the model,the coordinate attention(CA)channel attention mechanism is introduced,and the improved model is trained using a transfer learning strategy.The experiment shows that compared with the original model,the improved model CAM_qianyi with CA mechanism and transfer learning improves the accuracy by 0.82 percentage points,reaching 84.32%.At the same time,the accuracy of the improved model CAM_qianyi is also higher than that of the lightweight convolutional neural network ResNet18(82.54%)and the improved model CAM without transfer learning(73.65%).The improved model CAM_qianyi with transfer learning can accurately extract the characteristics of rice leaf disease and insect pest,effectively improving the efficiency and accuracy recognition.