Scholarlink

This paper aims to implement state-of-the-art (SOTA) models for real-time crack detection, specifically for use with Unmanned Aerial Vehicles (UAVs). The objective is to facilitate the easy and efficient detection of cracks, thereby enabling the protection of valuable cultural heritage buildings and the prevention of structural damage during adverse weather conditions such as monsoons or floods. In our specific case, YOLOv8 achieved a training time of 0.90 h and yielded impressive mean Average Precision (mAP) scores, including mAP@0.50 (bounding box) of 0.79, mAP@0.50–0.95 (bounding box) of 0.60, mAP@0.50 (mask) of 0.66, and mAP@0.50–0.95 (mask) of 0.26 on the validation dataset. Through our experiments, we determined that YOLOv8 is a superior model, exhibiting comparable speeds to YOLOv5 and accuracy levels on par with YOLOv7. Notably, we selected test images from various online sources and employed our models for crack detection, wherein YOLOv8 outperformed other models. These results firmly establish YOLOv8 as a highly effective and efficient solution for real-time crack detection.

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

Preserving Our Heritage: Buildings Deep Learning Solutions for Monitoring Cultural Heritage Structures Using Automated Crack Detection

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Rajdeep Chatterjee,
  • Biswaroop Nath

摘要

This paper aims to implement state-of-the-art (SOTA) models for real-time crack detection, specifically for use with Unmanned Aerial Vehicles (UAVs). The objective is to facilitate the easy and efficient detection of cracks, thereby enabling the protection of valuable cultural heritage buildings and the prevention of structural damage during adverse weather conditions such as monsoons or floods. In our specific case, YOLOv8 achieved a training time of 0.90 h and yielded impressive mean Average Precision (mAP) scores, including mAP@0.50 (bounding box) of 0.79, mAP@0.50–0.95 (bounding box) of 0.60, mAP@0.50 (mask) of 0.66, and mAP@0.50–0.95 (mask) of 0.26 on the validation dataset. Through our experiments, we determined that YOLOv8 is a superior model, exhibiting comparable speeds to YOLOv5 and accuracy levels on par with YOLOv7. Notably, we selected test images from various online sources and employed our models for crack detection, wherein YOLOv8 outperformed other models. These results firmly establish YOLOv8 as a highly effective and efficient solution for real-time crack detection.