With the rapid development of high-quality agriculture, the identification of crop diseases has become a hot research topic. Aiming at the research task of identifying various crop diseases, in this paper, ResNet18, channel attention mechanism and transfer learning are combined to achieve the goal of accurately identifying various diseases of various crops. The images of four crop diseases, corn, tomato, apple and grape, in PlantVillage of Tensorflow data set are used as training sets. The original ResNet18 is used for training and the identification effect is obtained; the combination of ResNet18 original model and transfer learning model is used for training and the identification effect is obtained; ResNet18 is proposed to combine with transfer learning model and channel attention mechanism for training. Finally, through the method of class activation map, the disease results are visualized, and the function of channel attention mechanism is further analyzed. The experimental results show that ResNet18 combined with channel attention mechanism and transfer learning model solves the problems of gradient explosion or disappearance, degradation and over-fitting, moreover achieves better extraction of feature information on pictures, which improves the identification accuracy of various crop diseases and the efficiency of training model. This method is suitable for crop disease identification tasks with complicated picture information and many situations. In this paper, many diseases of various crops are accurately identified, and identification methods suitable for complex crop diseases are put forward, which provides technical support for high-yield and high-quality agricultural development.