Scholarlink

<p>The increasing impact of global warming has intensified the frequency and severity of natural disasters, with flooding emerging as one of the most damaging events due to rising sea levels and unpredictable weather patterns. Accurate identification of flood-affected regions in ground-surveillance images is essential for rapid assessment and disaster response. Existing flood segmentation studies predominantly focus on satellite imagery, while methods targeting local or regional scenes often struggle with limited accuracy and generalization. This paper presents two deep learning–based segmentation architectures designed for ground-level flood detection, including a novel framework termed Flood-X. The proposed model utilizes an Xception-based encoder in conjunction with a lightweight decoder to achieve robust pixel-level segmentation. We further analyze the effect of three loss functions, Binary Cross-Entropy, Tversky Loss, and Log-Cosh Dice Loss, on model convergence and segmentation quality. Comparative experiments demonstrate that Flood-X achieves state-of-the-art performance, producing accurate segmentation masks on unseen real-world flood images and attaining an mIoU of 94% on the combined dataset, surpassing existing approaches in both efficiency and accuracy.</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

Optimizing surveillance efficiency with deep learning-driven flood segmentation

  • 期刊论文

  • 发表日期：

    2026年2月26日

查看全文
  • Shaurya Gupta,
  • Vinay Dubey,
  • Rahul Katarya

摘要

The increasing impact of global warming has intensified the frequency and severity of natural disasters, with flooding emerging as one of the most damaging events due to rising sea levels and unpredictable weather patterns. Accurate identification of flood-affected regions in ground-surveillance images is essential for rapid assessment and disaster response. Existing flood segmentation studies predominantly focus on satellite imagery, while methods targeting local or regional scenes often struggle with limited accuracy and generalization. This paper presents two deep learning–based segmentation architectures designed for ground-level flood detection, including a novel framework termed Flood-X. The proposed model utilizes an Xception-based encoder in conjunction with a lightweight decoder to achieve robust pixel-level segmentation. We further analyze the effect of three loss functions, Binary Cross-Entropy, Tversky Loss, and Log-Cosh Dice Loss, on model convergence and segmentation quality. Comparative experiments demonstrate that Flood-X achieves state-of-the-art performance, producing accurate segmentation masks on unseen real-world flood images and attaining an mIoU of 94% on the combined dataset, surpassing existing approaches in both efficiency and accuracy.