Scholarlink

<p>Natural disasters can cause extensive structural damage, necessitating rapid and reliable post-event assessment to support emergency response and recovery planning. Although several methods exist for pixel-level damage classification using post-disaster imagery, translating these outputs into meaningful, building-wise assessments remains challenging. Building-level damage prediction provides more interpretable insights, enabling a clearer estimation of the severity of impact on individual structures and a comprehensive understanding of the overall destruction. This information is crucial for quantifying damage magnitude and prioritizing relief operations. This paper proposes Damage Estimation U-Net (DE-U-Net), a deep learning framework designed to estimate structural damage across four classes: No Damage, Minor Damage, Major Damage, and Destroyed. The model is trained on the xBD dataset to learn representative damage patterns. DE-U-Net is developed by integrating a modified Siamese U-Net with a Damage Ratio Analyzer (DRA) algorithm for building-level damage conversion. The DRA algorithm comprises three components: (1) Connected Component Analysis (CCA) to transform pixel-level predictions into building-level predictions (2) size filtering to remove noise and eliminate small artifacts, and (3) a damage estimation module to compute the number of pixels corresponding to each damage class per building. Model performance is evaluated using standard metrics, including accuracy, precision, recall, and F1-score.</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

Remote sensing data analyzed by machine learning to predict structural changes

  • 期刊论文

  • 发表日期：

    2026年3月22日

查看全文
  • I. Sajitha,
  • Rakoth Kandan Sambandam,
  • Saju P. John

摘要

Natural disasters can cause extensive structural damage, necessitating rapid and reliable post-event assessment to support emergency response and recovery planning. Although several methods exist for pixel-level damage classification using post-disaster imagery, translating these outputs into meaningful, building-wise assessments remains challenging. Building-level damage prediction provides more interpretable insights, enabling a clearer estimation of the severity of impact on individual structures and a comprehensive understanding of the overall destruction. This information is crucial for quantifying damage magnitude and prioritizing relief operations. This paper proposes Damage Estimation U-Net (DE-U-Net), a deep learning framework designed to estimate structural damage across four classes: No Damage, Minor Damage, Major Damage, and Destroyed. The model is trained on the xBD dataset to learn representative damage patterns. DE-U-Net is developed by integrating a modified Siamese U-Net with a Damage Ratio Analyzer (DRA) algorithm for building-level damage conversion. The DRA algorithm comprises three components: (1) Connected Component Analysis (CCA) to transform pixel-level predictions into building-level predictions (2) size filtering to remove noise and eliminate small artifacts, and (3) a damage estimation module to compute the number of pixels corresponding to each damage class per building. Model performance is evaluated using standard metrics, including accuracy, precision, recall, and F1-score.