Scholarlink

<p>Remote sensing image change detection (CD) is crucial for environmental monitoring and urban development. Traditional deep learning methods, such as CNN-based approaches, are limited by their receptive fields, while Transformer-based methods suffer from quadratic computational complexity. We introduce spatial-reducing cross-attention change detection (SCACD), a novel remote sensing image CD method based on VMamba, achieving linear complexity and an expansive receptive field. Our approach employs the visual state space feedforward (VSFF) feature extraction network to capture multi-scale semantic features and utilizes the spatial-reducing cross-attention VSS (SCAV) module for efficient dual-temporal feature interaction. The cascaded convolutional attention (CCA) module further enhances multistage feature fusion. Experimental results on LEVIR-CD, SYSU-CD and WHU-CD datasets demonstrate superior performance, with SCACD outperforming state-of-the-art methods both visually and quantitatively. The source code and datasets are publicly available at <a href="https://github.com/QXL1220/SCACD.git">https://github.com/QXL1220/SCACD.git</a>.</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

Enhanced remote sensing change detection via VMamba and spatial-reducing cross-attention

  • 期刊论文

  • 发表日期：

    2026年3月21日

查看全文
  • Weidong Yan,
  • Mengtian Wang,
  • Chaosheng Zhu,
  • Delin Yu,
  • Zhihao Zou,
  • Tuo Xia

摘要

Remote sensing image change detection (CD) is crucial for environmental monitoring and urban development. Traditional deep learning methods, such as CNN-based approaches, are limited by their receptive fields, while Transformer-based methods suffer from quadratic computational complexity. We introduce spatial-reducing cross-attention change detection (SCACD), a novel remote sensing image CD method based on VMamba, achieving linear complexity and an expansive receptive field. Our approach employs the visual state space feedforward (VSFF) feature extraction network to capture multi-scale semantic features and utilizes the spatial-reducing cross-attention VSS (SCAV) module for efficient dual-temporal feature interaction. The cascaded convolutional attention (CCA) module further enhances multistage feature fusion. Experimental results on LEVIR-CD, SYSU-CD and WHU-CD datasets demonstrate superior performance, with SCACD outperforming state-of-the-art methods both visually and quantitatively. The source code and datasets are publicly available at https://github.com/QXL1220/SCACD.git.