Scholarlink

<p>Fortified settlements along the Great Wall are vital carriers of historical defense systems, yet systematic quantitative assessment of their preservation remains challenging. This study proposes an automated framework using high-resolution satellite imagery and the DeepLabV3+ deep learning model to segment and analyze the spatial preservation patterns of 75 settlements in Yuxian and along the Great Wall. The model achieved superior performance (IoU 0.8193) in extracting sparse, linear wall features compared to other architectures. Quantitative analysis revealed significant spatial heterogeneity: military fortresses exhibit higher preservation rates (74.26%) than civilian settlements (50.82%). Furthermore, a distinct directional bias was identified, with northern walls best preserved and eastern walls most degraded, likely driven by environmental factors and human encroachment. These findings demonstrate the efficacy of combining deep learning with remote sensing for the rapid, large-scale monitoring and risk assessment of extensive cultural heritage sites.</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

Satellite based DeepLabV3Plus semantic segmentation for fortress wall preservation assessment in Yuxian County

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月14日

查看全文
  • Chuanwen Luo,
  • Zikun Shang,
  • Yang You,
  • Zhe Zhang,
  • Ning Li,
  • Chunli Zhao,
  • Jian Yao,
  • Bo Zhang,
  • Shuqi Li

摘要

Fortified settlements along the Great Wall are vital carriers of historical defense systems, yet systematic quantitative assessment of their preservation remains challenging. This study proposes an automated framework using high-resolution satellite imagery and the DeepLabV3+ deep learning model to segment and analyze the spatial preservation patterns of 75 settlements in Yuxian and along the Great Wall. The model achieved superior performance (IoU 0.8193) in extracting sparse, linear wall features compared to other architectures. Quantitative analysis revealed significant spatial heterogeneity: military fortresses exhibit higher preservation rates (74.26%) than civilian settlements (50.82%). Furthermore, a distinct directional bias was identified, with northern walls best preserved and eastern walls most degraded, likely driven by environmental factors and human encroachment. These findings demonstrate the efficacy of combining deep learning with remote sensing for the rapid, large-scale monitoring and risk assessment of extensive cultural heritage sites.