Intelligent recognition of pavement crack based on RM-YOLOv5s algorithm

To prolong the service life of highways and enhance driving comfort and safety, detection of pavement cracks stands as a critical task in road monitoring. This article presents an innovative single-stage object detection approach based on You Only Look Once v5 Small (YOLOv5s) algorithm, aiming to enhance detection accuracy. Initially, the histogram equalization method processes the original data, enhancing the contrast between targets and the background. Then, a small-target detection layer is added to improve the network model's capability in detecting fine cracks. Subsequently, a RepViT visual module is incorporated to enhance the global feature representation capability of the network model, thereby strengthening its learning capacity. Additionally, the Soft-Non-Maximum Suppression (Soft-NMS) algorithm is adopted to replace Non-Maximum Suppression (NMS), which optimizes the suppression of redundant bounding boxes by jointly considering confidence scores and overlap degrees. Experimental results demonstrate that the enhanced RM-YOLOv5s model achieves mean Average Precision (mAP) of 94 %. Finally, the identified cracks are analyzed in terms of orientation, with their lengths and widths extracted. These improvements in pavement crack detection technology hold critical implications for the maintenance and repair of pavement infrastructure.