Research on bridge crack detection methods based on an improved YOLOv11n

This study introduces a bridge damage detection technique built upon an improved YOLO v11n architecture. The model integrates MSCA Attention within the backbone network, utilizing a spatial-channel cross-attention mechanism to capture cross-layer semantic relationships and strengthen multi-scale feature representation for defects. Concurrently, a C2f_DySnakeConv dynamic convolutional module is designed to adaptively fit the geometric characteristics of irregular defects such as surface cracks and corrosion on bridges. This enhances convergence speed, enabling high-precision bridge crack detection in complex environments. Experimental results indicate that the modified model gains a 4.6% higher mAP@0.5 and a 6.1% rise in mAPs over the original YOLOv11n, alongside substantially better detection performance for fine cracks and corrosion regions in complex environments.