Scholarlink

Three deep learning methodologies are proposed for enhanced landslide detection. An attention-augmented U-Net, utilizing Sentinel-2A imagery and Squeeze-and-Excitation modules, optimizes feature channel weighting to achieve precise landslide boundary segmentation. A multi-input U-Net integrates Sentinel-1A SAR, optical, and topographic data, enhancing detection accuracy in complex geomorphic settings. A Stacking ensemble, combining FCN, U-Net, and Attention U-Net, fuses spectral, topographic, and textural features to improve robustness and generalization. Applied to diverse terrains, these models demonstrate superior precision, recall, and F1-scores, validated through quantitative metrics and visual analyses, enabling reliable landslide mapping.

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

Deep Learning Approaches for Intelligent Landslide Detection

  • 章节

  • 发表日期：

    2025年

查看全文
  • Yi He

摘要

Three deep learning methodologies are proposed for enhanced landslide detection. An attention-augmented U-Net, utilizing Sentinel-2A imagery and Squeeze-and-Excitation modules, optimizes feature channel weighting to achieve precise landslide boundary segmentation. A multi-input U-Net integrates Sentinel-1A SAR, optical, and topographic data, enhancing detection accuracy in complex geomorphic settings. A Stacking ensemble, combining FCN, U-Net, and Attention U-Net, fuses spectral, topographic, and textural features to improve robustness and generalization. Applied to diverse terrains, these models demonstrate superior precision, recall, and F1-scores, validated through quantitative metrics and visual analyses, enabling reliable landslide mapping.