ES-ShuffleNetV2 Model for Maize Disease Recognition

In order to improve the accuracy and efficiency of maize disease recognition and take into account the practical requirements of portable equipment, a novel ES-ShuffleNetV2 maize disease lightweight recognition model was proposed in this paper. The new model is based on the ShuffleNetV2 model, adding the attention module SE after the first layer of convolution to enhance the model's attention to important features. In addition, by replacing the subsampling unit at each stage and the ReLU activation function of the base unit with the ELU function, a relatively small negative gradient inflow is allowed to make the gradient propagation smoother to accelerate the convergence rate of model training. Experimental results showed that the Accuracy of ES-ShuffleNetV2 in maize disease identification reached 99.82%, higher than the basic model's 96.05%, and the computational load remained at a relatively low level. Meanwhile, the Accuracy, F1-Score and other evaluation indexes were superior to other network models involved in the comparison. The effectiveness of the new model in maize disease identification was further verified.