YOLOv9-G-G-P: improved YOLOv9 pothole surface defect detection algorithm

Potholes are a major cause of traffic accidents, affecting driving stability and potentially leading to serious consequences such as vehicle loss of control and tire blowouts. Thus, timely detection and repair of potholes can significantly reduce the occurrence of accidents and improve road safety. This paper proposes a pothole surface defect detection algorithm, YOLOv9-G-G-P. First, GhostConv is introduced in the backbone to generate additional feature maps through low-cost linear transformations, significantly reducing model parameters and computational load. Second, the GhostInceptionV1 module is introduced to capture multi-scale features, enhancing the network’s understanding of image content and improving feature extraction capabilities. Lastly, due to small defects in the dataset, a P2 detection layer is added to improve the detection of small targets, recognizing finer image details. The improved model achieves an average precision of 80.5% and an accuracy of 86.2%, reflecting improvements of 1.5% and 1.2%, respectively, compared to the YOLOv9 model, while reducing model parameters by 30.3%. By integrating advanced detection algorithms like YOLOv9, an intelligent road monitoring system can be established for real-time pothole monitoring and early warnings, helping to reduce accidents and vehicle damage caused by potholes.