YOLOv9-GHP: Pothole Surface Defect Detection Algorithm Based on YOLOv9

Road potholes represent a significant hazard in the field of traffic safety, and their potential dangers cannot be ignored. These potholes not only pose serious threats to moving vehicles, leading to mechanical failures such as abnormal tire wear and suspension system damage, but they also undermine the safety of drivers and passengers. This paper proposes a pothole surface defect detection algorithm, YOLOv9-GHP. Firstly, the C3Ghost module is introduced in the backbone, which not only reduces the number of parameters but also enhances multi-scale feature fusion capabilities. This allows the model to capture semantic information at different levels, improving its performance. Secondly, the HWD-ADown downsampling is introduced, which increases the network depth, extends the receptive field, and enhances the ability to extract deep features of defects. Lastly, the addition of the P2 detection layer enables the model to capture small defects with high accuracy, reducing missed and false detections and improving model precision. The improved model achieves an average precision and recall rate of 80.8% and 71.6%, respectively. The integrated YOLOv9 algorithm endows the monitoring system with the capability to process image data in real-time and identify road potholes, providing timely warning information to road maintenance teams.