Scholarlink

<p>Prolonged exposure to natural factors and human activities has caused severe brick missing damages in many Great Wall defensive forts, weakening their load-bearing structures. Given the Great Wall’s vast scale, remote locations, and complex terrain, there is an urgent need for a method to quickly identify and locate such damages to support daily monitoring and maintenance. This study proposes a computer vision-based two-phase automatic damage detection and localization method for ancient city walls. In phase one, an Improved-YOLOv5n object detection network, trained on 1197 UAV images, integrates attention mechanisms and knowledge distillation to enhance small target detection, achieving a mean average precision of 74.5%. In phase two, a genetic algorithm-optimized multi-threshold OTSU segmentation and image processing are used to localize damages and extract edge locations, aiding subsequent modeling. The findings of this study can provide a time-efficient, high-accuracy and non-destructive solution for routine structural safety assessments of ancient city walls.</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

Automatic damage detection and localization of ancient city walls—a case study of the Great Wall

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年5月15日

查看全文
  • Shiyu Ma,
  • Qing Chun,
  • Chengwen Zhang,
  • Dongqing Li,
  • Fei Zhai,
  • Yu Yuan

摘要

Prolonged exposure to natural factors and human activities has caused severe brick missing damages in many Great Wall defensive forts, weakening their load-bearing structures. Given the Great Wall’s vast scale, remote locations, and complex terrain, there is an urgent need for a method to quickly identify and locate such damages to support daily monitoring and maintenance. This study proposes a computer vision-based two-phase automatic damage detection and localization method for ancient city walls. In phase one, an Improved-YOLOv5n object detection network, trained on 1197 UAV images, integrates attention mechanisms and knowledge distillation to enhance small target detection, achieving a mean average precision of 74.5%. In phase two, a genetic algorithm-optimized multi-threshold OTSU segmentation and image processing are used to localize damages and extract edge locations, aiding subsequent modeling. The findings of this study can provide a time-efficient, high-accuracy and non-destructive solution for routine structural safety assessments of ancient city walls.