Research on crop diseases classification model based on MobileNet.

To address the problems of large size and low-recognition accuracy of the Convolutional Neural Networks (CNNs) for crop disease recognition, this paper proposes an improved model of MobileNet called MobileNet-LR-SE. The model firstly uses the LR structure to reverse the feature information and get the new feature information. These together with the original feature information form the input to the feature fusion layer. The LR structure introduces residual connection and uses the Leaky ReLU activation function. Secondly, it embeds the SE module to complete the final image classification. The LR structure solves the problem of ignoring negative feature information in the training process of ordinary neural networks. The SE module improves the attention of the network model to the useful channels. Experiments show that the MobileNet-LR-SE model has a high-accuracy rate when evaluated on the two crop disease data sets, with the number of network parameters of only 2.01M.