A method for identifying crop diseases based on IAlexNet model

Abstract

With the decrease of farmers and the urgent needs of agricultural modernization, deep learning becomes a novel and effective way to identify crop diseases in modern agriculture. For the problems about low accuracy and complexity of models, a light-weight disease recognition model based on AlexNet is proposed, which is called IAlexNet. The large convolution kernel is replaced by several small convolution kernels to reduce the network parameters, and the SE-Net is introduced to increase the weight of effective information. Besides, the dataset uses the pathological image datasets of apple leaves published on AI studio of the Paddlepaddle. The experiment results show that the recognition accuracy is 97.16%, which is 1.95% higher than AlexNet model. In addition, the parameters of IAlexNet model are reduced by 59.11%, and the training time is reduced by 20.33%, which is verify the new proposed model is feasible and effective.