C-YOLO: An Attention-Enhanced Framework for Road Crack Detection

Timely and accurate detection of road surface cracks is essential for infrastructure maintenance and traffic safety. Yet, many existing detection algorithms struggle to effectively capture cracks of varying scales and shapes, especially when faced with complex textures and noise from real-world road scenes. To overcome these challenges, this study presents an enhanced detection framework, termed C-YOLO, which is specifically optimized for crack identification. The model incorporates a bidirectional feature pyramid network (BiFPN) to strengthen the aggregation of features across different resolutions, addressing the shortcomings of earlier models in multi-level feature representation. To improve the model’s sensitivity to subtle crack features, particularly under poor lighting or interference, an Efficient Channel Attention (ECA) mechanism is embedded to emphasize key feature channels. Furthermore, a Convolutional Block Attention Module (CBAM) is applied to refine both spatial and channel-wise focus, thereby enhancing the network's ability to differentiate fine cracks from background textures. Experimental evaluations indicate that C-YOLO outperforms the baseline by 4.6% in mAP, confirming its effectiveness in detecting cracks under diverse and challenging conditions.