As industrial automation and intelligence advance progressively, the necessity for enhanced precision in bearing defect identification and detection is on the rise. Deep learning intelligent algorithms facilitate the convenient and sophisticated recognition of bearing defects, addressing current demands effectively. However, prevalent models are encumbered by substantial parameter counts and suffer from suboptimal accuracy, compromising the reliability of defect identification processes. This study investigates the problem of bearing defect detection by employing a YOLOv5-CBAM-based methodology for recognition purposes. The integration of the CBAM (Convolutional Block Attention Module) attention mechanism into the architecture of the YOLOv5s backbone network has facilitated the development of an enhanced YOLOv5 model, specifically optimized for the purpose of bearing defect detection. The analysis of experimental outcomes indicates that the enhanced algorithm introduced in this study achieves a significant reduction in model complexity, with the total number of parameters reduced to 6,709,997. This represents a decrease of 316,310 parameters in comparison to the YOLOv5s algorithm model. The mean Average Precision at 50% (mAP50) achieved an accuracy of 0.791, representing a 0.6% enhancement relative to the accuracy of the original model.