Research on Road Defect Detection Based on Improved YOLOv5s

In common road traffic scenarios, the presence of road defects represented by cracks and potholes poses safety hazards to people's travel. Currently, manual methods are still predominant in road defect detection, which not only consumes time and effort but also tends to result in missed detections and false alarms. To address this issue, this paper proposes the G-CA-YOLOv5s model by improving the YOLOv5s model, replacing the original backbone network with a more lightweight Ghost Net, and introducing the Coordinate Attention (CA) to address the issue of reduced feature extraction caused by the change in the backbone network. The experimental results show that compared to the original model, the G-CA-YOLOv5s model reduces the parameter count by 21.23%, decreases computation by 8.441GB, and reduces model size by 20.30%, with only a 1.8% decrease in mAP score reflecting model detection accuracy. The improved model's network complexity is further reduced, significantly lowering the computational requirements for platforms and enabling effective deployment on hardware platforms with limited GPU computing power such as mobile devices.