Abstract: Targeting the problem of low accuracy of one-dimensional convolutional neural network and two-dimensional convolutional neural network in wind turbine bearing fault diagnosis,the paper combines the one-dimensional original vibration signal with twodimensional time-frequency map,and proposes a rolling bearing fault diagnosis method based on CBAM-InceptionV2-two-stream CNN. Firstly,the original vibration signal is converted into the one-dimensional data and two-dimensional time-frequency map with fast Fourier transformation（FFT）and wavelet transformation. Then a CBAM-InceptionV2-two-stream CNN model is established.Finally,the extracted double level feature information is fused and input to SoftMax to complete the fault classification. The experimental results show that the proposed model can significantly enhance the accuracy of the bearing fault diagnosis.