Rolling bearings are widely used in coal mine equipment, real-time monitoring of bearing status and intelligent diagnosis of bearing fault is very important to the safe and efficient mining of coal mine. In order to solve the problem of insufficient representation ability about fault diagnosis model, a multi-scale segmentation attention-based residual network is proposed for the fault diagnosis of rolling bearings, which can fully and accurately extract vibration signal features to realize intelligent diagnosis. For time-frequency images of vibration signals, residual networks was used for feature extraction. Furthermore, the pyramid split attention mechanism was combined to optimize the feature selection to construct the intelligent diagnosis model for bearing. The proposed method has been applied to the task of fault diagnosis on SKF6205 bearing, and the experimental results show its superior diagnostic performance.