Research on Agricultural Disease Recognition Based on Improved MobileNetV2

Aiming at the problems of low recognition accuracy and low feature extraction ability of existing lightweight neural networks in agricultural disease detection, this paper proposes a lightweight network model to improve MobileNetV2. First, the residual block is redesigned in the network. The Shuffle Attention (SA) mechanism is introduced, which effectively enhances the feature extraction ability of the model for space and channel features through channel shuffling, and improves the model's attention to essential features; at the same time, the gated residual connection technique is adopted, and two learnable parameters are introduced to improve the model's flexibility and enhance the feature extraction ability; Finally, the network structure is optimized by removing the dropout layer and introducing Swish activation function. The experimental results show that the improved model improves the accuracy by 1.5% compared to the original MobileNetV2 network, and the feature extraction ability of the system is significantly enhanced by the introduction of gated residuals. The feasibility of this project is verified by deploying it on edge devices.