Scholarlink

<p>Remote sensing change detection (RSCD) aims to identify and extract surface changes from multi-temporal remote sensing images. However, existing CNN-based methods often face challenges, such as heavy computational costs and limited deployment feasibility. To overcome these limitations, we propose MCLNet, a lightweight and high-performance RSCD network that achieves a better balance between accuracy and efficiency. The model adopts a dual-encoder architecture to enhance feature representation and employs multi-level dense connections with a differential fusion strategy to improve change sensitivity while reducing redundant parameters. Additionally, a channel attention mechanism is incorporated to strengthen the extraction of subtle local changes. Experimental results on two benchmark RSCD datasets demonstrate that MCLNet not only achieves state-of-the-art accuracy with the lowest computational cost (6.12G) and a relatively small number of parameters (7.82&#xa0;M), but also exhibits strong generalization and deployment potential for large-scale environmental monitoring, land-use analysis, and disaster assessment. Overall, MCLNet provides a practical and scalable solution for efficient change detection in real-world remote sensing applications.</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

MCLNet: a lightweight model based on multi-level dense connections and channel attention for remote sensing change detection

  • 期刊论文

  • 发表日期：

    2025年12月22日

查看全文
  • Shoubin Wang,
  • Pengcheng Gao,
  • Guili Peng,
  • Yuan Zhou,
  • Kun Li,
  • Zefeng Ding,
  • Xinchang Fang,
  • Zimeng Gao

摘要

Remote sensing change detection (RSCD) aims to identify and extract surface changes from multi-temporal remote sensing images. However, existing CNN-based methods often face challenges, such as heavy computational costs and limited deployment feasibility. To overcome these limitations, we propose MCLNet, a lightweight and high-performance RSCD network that achieves a better balance between accuracy and efficiency. The model adopts a dual-encoder architecture to enhance feature representation and employs multi-level dense connections with a differential fusion strategy to improve change sensitivity while reducing redundant parameters. Additionally, a channel attention mechanism is incorporated to strengthen the extraction of subtle local changes. Experimental results on two benchmark RSCD datasets demonstrate that MCLNet not only achieves state-of-the-art accuracy with the lowest computational cost (6.12G) and a relatively small number of parameters (7.82 M), but also exhibits strong generalization and deployment potential for large-scale environmental monitoring, land-use analysis, and disaster assessment. Overall, MCLNet provides a practical and scalable solution for efficient change detection in real-world remote sensing applications.