Aiming at the lack of adaptability as the traditional shallow bearing fault diagnosis method relying on artificial feature extraction and diagnosis expertise, a fault diagnosis method based on deep convolutional neural network is proposed to recognize twodimensional shapes. Firstly, in order to fully display the fault characteristic information of rolling bearing, the twodimensional time spectrum of rolling bearing vibration time series is obtained by using the shorttime Fourier transform. Secondly, different fault features are extracted by convolutional neural network adaptively. Finally, Softmax classifier is used to output the diagnosis results to realize bearing fault diagnosis. The results show that the accuracy of the measured bearing fault diagnosis is up to 999%, proves that the proposed method has a good generalization performance and feasibility.