Abstract: Aiming at the problems of complex working conditions, insufficient feature extraction and low fault diagnosis accuracy with a small dataset, a rolling bearing fault diagnosis model based on Markov transfer field (MTF), convolutional block attention module (CBAM) and improved residual neural network (IResNet) is proposed. Firstly, the MTF algorithm retains the time-dependent characteristics in the one-dimensional vibration signals and generates a two-dimensional image. Secondly, CBAM is used to capture the key features of the image and dynamically learns the relationship between the features at different scales. Thirdly, the IResNet enhances the nonlinear expression ability of the network. Finally, the MTF-CBAM-IResNet model is constructed for fault diagnosis. The experimental results show that the average accuracy of the model reaches 99.29% under variable operating conditions, and 99.05% and 97.67% under different scales of small samples, which verifies the generalization performance and diagnostic effect of the model.