Scholarlink

Dam banks deformation monitoring enables the early detection of potential hazards and the way based on video surveillance provides a simple, convenient, low-cost, and non-contact monitoring manner compared with the other contact-type monitoring devices. While, it faces many challenges: large background changes, the coexistence of fast and slow changes, and the lack of a dam dataset. To emphasis these issues, we build a dam segmentation dataset (DamSeg) and a lightweight dam semantic segmentation network (Dam-YOLO), so that the change detection can efficiently focus on dam bank regions. Then, we adopt Flow-CDNet to simultaneously detect fast and slow changes. Finally, Dam-CDNet is presented, which integrates the Dam-YOLO trained on DamSeg and Flow-CDNet, to achieve accurate dam deformation detection. Experiments show Dam-YOLO achieves the best overall speed-accuracy balance, and quantitative tests on the self-built DamChange dataset show Dam-CDNet further improves dam deformation monitoring compared to Flow-CDNet.

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

Dam Deformation Monitoring via Dam Segmentation and Fast and Slow Change Detection

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Chenxu Wei,
  • Haoxuan Li,
  • Boyuan An,
  • Lingyan Ran,
  • Baosen Zhang,
  • Junrui Liu,
  • Xiuwei Zhang

摘要

Dam banks deformation monitoring enables the early detection of potential hazards and the way based on video surveillance provides a simple, convenient, low-cost, and non-contact monitoring manner compared with the other contact-type monitoring devices. While, it faces many challenges: large background changes, the coexistence of fast and slow changes, and the lack of a dam dataset. To emphasis these issues, we build a dam segmentation dataset (DamSeg) and a lightweight dam semantic segmentation network (Dam-YOLO), so that the change detection can efficiently focus on dam bank regions. Then, we adopt Flow-CDNet to simultaneously detect fast and slow changes. Finally, Dam-CDNet is presented, which integrates the Dam-YOLO trained on DamSeg and Flow-CDNet, to achieve accurate dam deformation detection. Experiments show Dam-YOLO achieves the best overall speed-accuracy balance, and quantitative tests on the self-built DamChange dataset show Dam-CDNet further improves dam deformation monitoring compared to Flow-CDNet.