Image recognition of soybean leaf disease based on bilateral branching network BBN

Jianming Sun, Xuyao Hao, Mengxin Zhao

Abstract: In recent years, with the rapid growth of China's economy, environmental pollution has become more and more serious. Fog and haze weather, water pollution and other factors have led to the occurrence of soybean diseases, seriously affecting the quality and yield of soybean. With the development of computer vision technology and deep learning and gradually applied to the development of agriculture, it is possible to use computers to intelligently diagnose crop diseases. Due to the different incidence of different types of soybean diseases in different regions and at different times, the long tail distribution problem of soybean leaf disease datasets is caused, which leads to the overall performance of disease recognition models. To solve this problem, BBN_IC_MobileNetV3 network model is proposed. The new network is based on MobileNetV3. Firstly, the multi-scale feature extraction module is used to replace the first 3×3 convolution layer of the original network to improve the feature extraction ability of the network for different areas of disease spots. Secondly, the CA attention mechanism is used to replace the SE attention mechanism of the original network to distinguish between the target and background. Finally, the improved MobileNetV3 network is combined with the BBN network to improve the recognition accuracy of the tail data by modeling the tail data, so as to improve the overall performance of the model. Experiments show that the overall disease recognition rate of the BBN_IC_MobileNetV3 network model reaches 95.882%, and the recognition accuracy of the tail data is close to 90%, which has been improved compared with the original network.