In order to solve the problem of artificial characteristics of bearing fault diagnosis, this paper puts forward the fault diagnosis method of the time frequency analysis and VGG19 network migration learning. First, the normal state, the internal ring fault, the outer ring fault and the sliding fault of the rolling body are converted to the frequency sample diagram, and then the spectrum kurtosis graph is generated from the above data. Secondly, the full connection layer in the VGG19 network model is replaced and fine-tuning. Finally, the convolutional neural transfer learning network is used to recognize and classify bearing faults through network parameter tuning.The results show that the classification accuracy of time-frequency sample graph for rolling bearing fault diagnosis in the experiment is 5.42% higher than that of spectral kurtosis graph, which verifies the validity of the application of time-frequency analysis and VGG19 transfer learning in signal processing.In addition,Transfer learning can solve the problem of fault diagnosis of small samples.