Identification of maize disease based on transfer learning

To reduce the time and energy needed by agricultural experts to evaluate maize diseases and improve the accuracy of image recognition of maize diseases, considering that the maize disease dataset is a small sample dataset, a method of image recognition of maize disease based on transfer learning is proposed, which is to fine tune the pre-trained convolutional neural network model. This paper compares and analyzes the experimental results of training only the last layer of Inception-v3 / Inception-v4 model and training all parameters of Inception-v3 / Inception-v4 model, and compares the results with the initial training VGG16, AlexNet and other models. The experimental results show that the method of transfer learning can reduce the training time of network, and the trained deep convolutional neural network can recognize maize diseases.