At present, the identification and diagnosis of some typical crop leaf diseases mainly rely on artificial method, which is time-consuming and laborious. Aimed at the diagnosis of common leaf diseases of five typical crops of soybean, cotton, rice, wheat and maize, a recognition method of typical leaf diseases of crops based on convolution neural network was proposed. Leaf disease images of typical crops were collected from the Plantvillage database and some other sites, and these images were pretreated to build a database of 12 836 sheets. Referring to AlexNet framework, an eight-layer convolutional neural network was constructed, and the transfer learning training network was adopted. Finally, the recognition accuracy and loss value of the network were verified by the test set. The performance of different convolutional neural networks was analyzed. The experimental results showed that the algorithm performed well in identifying typical crop leaf diseases. Under the transfer learning mode, with the learning rate of 0.001, the recognition accuracy of the algorithm in the training set was about 99.47%, and about 96.18% in the test set.