Regarding the problems of fault data scarcity, complex signal processing, and high dependence on professional knowledge in traditional motor bearing fault diagnosis methods, this paper presents a few-shot fault diagnosis method of motor bearing by combining Markov transition field and Swin Transformer neural network. Firstly, based on the theory of compressed sensing, the original vibration signal of the motor bearing is compressed sampling and reconstructed to realize the expansion generation of limited fault data. Then, to avoid complex signal processing, the Markov transition field method is introduced to convert one-dimensional data in the time domain into two-dimensional images. Finally, Swin Transformer network model is introduced and combined with two-dimensional visualization images to realize the classification and identification of different bearing fault types. The results show that the proposed diagnosis method can effectively classify different bearing fault types, and the fault recognition accuracy on three standard datasets can reach 96.49%, 99.52%, and 96.07 respectively.