Research on Rail Surface Crack Detection Based on Improved YOLOv5

With the rapid development of world's railroad infrastructure, timely detection of track surface crack has become critical for ensuring operational safety. These cracks primarily arise from prolonged train movements, temperature fluctuations, and natural disasters. Current detection methods suffer from limitations such as low precision, poor recall rates, and inadequate performance in small-object detection. To address these challenges, this paper proposes an enhanced YOLOv5s-CBE model by integrating three key improvements : Dual attention enhancement: 1) The Convolutional Block Attention Module (CBAM) is embedded into both the backbone C3 module and the connection between C3 and SPPF, enabling simultaneous channel and spatial feature refinement to strengthen critical crack information extraction. 2) Multi-scale feature fusion: A Bidirectional Feature Pyramid Network (BiFPN) is adopted to optimize cross-scale feature aggregation, significantly improving detection capability for small and multi-scale cracks. 3) Loss function optimization: The original CIoU loss is replaced with EIoU, which explicitly incorporates centroid distance, aspect ratio, and width-height discrepancies between predicted and ground-truth boxes, thereby enhancing anchor regression accuracy .Experimental results demonstrate that the proposed YOLOv5s-CBE achieves a mean average precision (mAP@0.5) of 76.3% on the rail surface crack dataset, representing improvements of 7.7% over the baseline YOLOv5s, respectively. Furthermore, the model exhibits lightweight characteristics, with reduced computational complexity and parameter size, effectively mitigating issues of false positives and missed detections in real-world scenarios. This advancement provides a robust solution for automated and high-precision rail surface crack inspection.