Research on Expressway Pavement Crack Detection based on Improved YOLOv5s

Abstract

In order to address the issues of missed detection, false detection, and low accuracy of current road cracks, we propose a road crack recognition model based on improved YOLOv5. Firstly, add a CBAM attention module to the backbone network to enhance feature extraction capabilities; Then, a weighted bidirectional feature pyramid (BiFPN) is incorporated into the model for multi-scale feature fusion, replacing the traditional feature pyramid (FPN)+pixel aggregation network (PAN) structure to enhance feature fusion. The experimental results indicate that the improved model outperforms the traditional YOLOV5 model in terms of mAP@0.5 By 17.3%, the improved YOLOv5 algorithm performs well in detecting road cracks and can quickly and accurately identify and locate cracks on the road.