The rolling bearing of a wind turbine has problems of weak fault characteristics, difficult extraction and low diagnosis efficiency. To solve these problems we propose a fault diagnosis algorithm based on an improved Convolution Neural Network (CNN). The structure of the CNN model is improved, a new convolution layer is added in front of the full connection layer, the deep features of the signal are excavated to improve the generalization ability of the model, the convolution layer data are standardized, and the stochastic gradient descent with momentum is used to speed up the training speed. The working principle of the improved CNN is introduced in detail, and the flow chart of fault diagnosis with improved CNN is given. Finally, the data of a rolling bearing database at Case Western Reserve University is used to verify the method, and proves that this method does not need to extract the fault features of the signal in advance, and can directly achieve fault feature extraction and fault identification of bearings, and the diagnosis rate is high. This work is supported by the National Natural Science Foundation of China (No. 61573167) and the Fundamental Research Funds for the Central Universities (No. JUSRP51510).