Comparative Analysis of YOLOv5 and YOLOv7 for UAV-based Automated Road Damage Detection

This study introduces an innovative approach for identifying road damage using unmanned aerial vehicle (UAV) imagery and advanced deep learning models. The research undertakes a rigorous and critical comparative analysis of the latest iterations of the You Only Look Once (YOLO) algorithm, specifically versions 5 and 7. This comprehensive review work is crucial for systematically evaluating their performance and practical utility in automating the detection and classification of various road damage types, including longitudinal cracks, transverse cracks, alligator cracks, and potholes. High-resolution images are collected via UAVs to provide comprehensive coverage of extensive road networks, enabling efficient and thorough inspection. The YOLO models are trained on a curated dataset of annotated UAV images, capturing diverse damage patterns across various environmental conditions. The performance of YOLO v5 and v7 models is evaluated in terms of precision, speed, and reliability in detecting damage. The results indicate that while both models are effective, YOLOv7 outperforms YOLOv5. Specifically, YOLOv7 achieves a higher average precision of 99.35% compared to YOLOv5's 98.75%, demonstrating YOLOv7's superior accuracy in identifying road damage. Additionally, YOLOv7 exhibits a higher overall accuracy of 99.35% compared to YOLOv5's 98.90%. This enhanced performance of YOLOv7 translates to fewer false positives and more reliable detection of actual damage, thereby providing valuable insights and a well-substantiated basis for decision-making regarding the deployment of advanced object detection systems in real-world road infrastructure management.