Scholarlink

<p>Advancements in deep learning and computer vision have significantly improved object detection, particularly in infrastructure assessment. Accurate identification of various road surface distresses, such as cracks, potholes, rutting, and patchwork, is crucial for maintaining road performance and ensuring public safety. Traditional inspection methods are time-consuming, subjective, and inefficient for large-scale applications. This study compares the performance of multiple versions of the YOLO object detection algorithm YOLOv3, YOLOv5, YOLOv7, and YOLOv8 in detecting and classifying flexible pavement distresses from different angle images. This study shows that YOLOv8 achieved the highest accuracy, with an overall mean Average Precision (mAP) of 82.6%, surpassing YOLOv7 (70.0%) and YOLOv5/v3 (around 60.0%). Specifically, YOLOv8 demonstrated exceptional performance in detecting rutting (91.7% mAP), alligator cracking (85.7% mAP), and potholes (81.2% mAP), while also boasting a faster inference time of 0.018s per image, making it suitable for real-time applications. These findings underscore the potential of integrating efficient and automated distress detection with YOLO-based deep learning frameworks. This study contributes to the ongoing evolution of intelligent road maintenance systems by providing insights into the efficacy of advanced YOLO models for road surface distress detection. This study contributes not by proposing a new algorithm but by delivering a practical, real-world comparison of YOLO versions for all types of pavement distress detection, guiding model selection for scalable, real-time deployment in infrastructure monitoring.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Optimizing YOLO models for high-accuracy automated detection and classification of road surface distresses

  • 期刊论文

  • 发表日期：

    2025年7月28日

查看全文
  • M. Manjusha,
  • V. Sunitha

摘要

Advancements in deep learning and computer vision have significantly improved object detection, particularly in infrastructure assessment. Accurate identification of various road surface distresses, such as cracks, potholes, rutting, and patchwork, is crucial for maintaining road performance and ensuring public safety. Traditional inspection methods are time-consuming, subjective, and inefficient for large-scale applications. This study compares the performance of multiple versions of the YOLO object detection algorithm YOLOv3, YOLOv5, YOLOv7, and YOLOv8 in detecting and classifying flexible pavement distresses from different angle images. This study shows that YOLOv8 achieved the highest accuracy, with an overall mean Average Precision (mAP) of 82.6%, surpassing YOLOv7 (70.0%) and YOLOv5/v3 (around 60.0%). Specifically, YOLOv8 demonstrated exceptional performance in detecting rutting (91.7% mAP), alligator cracking (85.7% mAP), and potholes (81.2% mAP), while also boasting a faster inference time of 0.018s per image, making it suitable for real-time applications. These findings underscore the potential of integrating efficient and automated distress detection with YOLO-based deep learning frameworks. This study contributes to the ongoing evolution of intelligent road maintenance systems by providing insights into the efficacy of advanced YOLO models for road surface distress detection. This study contributes not by proposing a new algorithm but by delivering a practical, real-world comparison of YOLO versions for all types of pavement distress detection, guiding model selection for scalable, real-time deployment in infrastructure monitoring.