Scholarlink

<p>Slum detection has emerged as a critical area of research in urban studies and remote sensing, driven by the need for effective urban planning, resource allocation, and policy intervention during disaster management. Detecting slums from satellite imagery presents a significant challenge due to variations in building materials, density, spatial organization, and local environmental factors, making it difficult to establish a universal detection approach. This study makes use of all bands of Sentinel-2 up to 20&#xa0;m along with four spectral indices and an ignore mask to detect slum pockets in Mumbai. An attention mechanism is incorporated into U-Net architectures to enhance feature representation and improve classification performance. The model was further evaluated for generalization and transferability on a patch from the Jakarta dataset, where it demonstrated strong performance in slum classification. This work can be used by government bodies to continuously track and monitor the growth of slum areas and assist decision-makers in making informed decisions. Our proposed method with the attention module focuses on the most relevant spatial and contextual information within the image, strengthening discriminative feature learning and achieving superior results for slum detection.</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

Slum Detection Using U-Net Attention Gate on Mid-Resolution Satellite Imagery

  • 期刊论文

  • 发表日期：

    2026年6月18日

查看全文
  • Archana G. Pai,
  • Aaron Augustine,
  • Rupesh Kumar Yadav Mediboyina,
  • Ankit Jha,
  • Krishna M. Buddhiraju,
  • Surya S. Durbha

摘要

Slum detection has emerged as a critical area of research in urban studies and remote sensing, driven by the need for effective urban planning, resource allocation, and policy intervention during disaster management. Detecting slums from satellite imagery presents a significant challenge due to variations in building materials, density, spatial organization, and local environmental factors, making it difficult to establish a universal detection approach. This study makes use of all bands of Sentinel-2 up to 20 m along with four spectral indices and an ignore mask to detect slum pockets in Mumbai. An attention mechanism is incorporated into U-Net architectures to enhance feature representation and improve classification performance. The model was further evaluated for generalization and transferability on a patch from the Jakarta dataset, where it demonstrated strong performance in slum classification. This work can be used by government bodies to continuously track and monitor the growth of slum areas and assist decision-makers in making informed decisions. Our proposed method with the attention module focuses on the most relevant spatial and contextual information within the image, strengthening discriminative feature learning and achieving superior results for slum detection.