Concrete Bridge Crack Detection Based on YOLO v8s in Complex Background

A crack detection algorithm based on improved YOLO v8s is proposed to address the problem of poor crack detection accuracy of concrete bridge in complex background. This method utilizes an improved MA-ECA channel attention module to construct C2f-MA for mining more texture information in the feature map, further focusing the network on crack features, suppressing irrelevant background information, and improving the crack detection performance. Additionally, a small target layer with a size of 160 × 160 is added to precisely locate and identify small targets using rich semantic information contained in shallow features, thereby reducing the loss of small crack features and improving the accuracy of crack detection. Finally, experimental evaluations were conducted on a self-made concrete bridge crack dataset. The proposed algorithm showed the characteristics of high accuracy, few parameters, and fast speed compared to currently classical object detection algorithms in complex background. Moreover, the proposed algorithm demonstrated a 2.0% improvement in detection accuracy compared to the baseline network YOLO v8s.