In view of the problem that the accuracy of existing fault diagnosis methods is not high for bearing faults with different degrees of small classification,a fault diagnosis method based on GWO-CMFH and improved ResNet was proposed.For the same type of fault with different degrees,a CMFH filtering method based on GWO adaptive optimization structural element parameters was proposed to enhance the impulse fault characteristics of the vibration signal and suppress the background noise.Continuous wavelet transform was used to convert the filtered signal into a 2D time-frequency map.Finally,an improved residual network model based on the mixed attention mechanism was proposed to improve the bearing fault diagnosis accuracy.The verification experiments were carried out on Western Reserve University,Southeast University and the selected bearing datasets,and the diagnostic accuracy rates of small classification faults of different degrees reached 99.73%,98.12% and 99.07%,respectively.It shows that the proposed method has good anti-noise and robustness,and it can improve the fault diagnosis effect of rolling bearings with different degrees fault.