Crack Detection Based on an Enhanced YOLO Method

Identifying pavement defects is a crucial endeavor for maintaining road safety. Traditional machine vision methods struggle to attain satisfactory detection results due to the intricate and varied nature of pavement defects and the differing sizes of defect areas. This paper proposes a data augmentation-based target detection algorithm utilizing YOLOV11 (You Only Look Once) to ensure the safe operation of traffic, given the limited existing data on road pothole defects. The research background and significance of algorithms for detecting pavement diseases are presented. The enhancements and benefits of the YOLOV11 model are examined next. Ultimately, utilizing experimental data augmentation techniques, the average accuracy of the most efficacious method can be enhanced by 8 percentage points through various combinations of data augmentation. The proposed enhanced road defect dataset is subsequently compared with alternative methods. The findings demonstrate that this method possesses overall benefits, evidenced by enhancements in mAP, regression rate, and recall rate. These studies offer viable solutions for pavement imperfections and establish a robust foundation for the design and advancement of defect detection.