PavD-YOLOv9: An Enhanced Deep Learning Method for Accurate Pavement Distress Detection and Classification

Pavement distress detection is essential for maintaining road stability and ensuring corrective action. Conventional techniques for keeping an eye on road conditions are frequently laborious and prone to human mistakes. PavD-YOLOv9, an improved deep learning model, is proposed in this work to improve and automate pavement distress detection and classification. By training the model on a diverse set of road surface images, the proposed model can accurately identify various distresses such as cracking, potholes, rutting, patches, and so on. The model’s architecture, optimized with custom layers and hyperparameters, improves its performance in challenging real-world conditions. Experimental results show a notable improvement of 4% in precision, mAP, F1-score, and 2% improvement in recall, and detection speed compared to the default YOLOv9c model, making it a reliable tool for real-time pavement analysis. The proposed model achieved a precision of 95%, recall 90%, mAP 88%, and F1-score of 74%. By utilizing YOLOv9, pavement distress monitoring becomes more efficient, cost-effective, and sustainable, leading to better road safety and infrastructure management.