Scholarlink

<p>With the rapid development of artificial intelligence and remote sensing imaging, change detection (CD) has become increasingly critical in applications such as urban planning, disaster assessment, and environmental monitoring. However, existing methods still struggle to effectively fuse bi-temporal features and accurately identify change regions due to challenges such as scale variation and semantic drift. To address these issues, this paper proposes a novel deep learning-based framework, the Change Detection Feature Fusion Network (CDFF-Net), which enhances the representational power of fused features and improves change detection performance. Built upon the U-Net backbone, CDFF-Net introduces two key modules: (1) the Change Guidance Module (CGM), which dynamically recalibrates deep features using change priors to improve sensitivity to true changes; and (2) the Parallel Partial Convolutional Attention (PPCA) module, which maintains the integrity of salient features through multiscale selective enhancement during bi-temporal fusion. Extensive experiments on three benchmark remote sensing datasets Sun Yat-sen University Change Detection Dataset (SYSU-CD), Large-scale Building Change Detection Dataset from LEVIR (LEVIR-CD), and Wuhan University Change Detection Dataset (WHU-CD) demonstrate that CDFF-Net achieves F1-scores of 81.66%, 92.14%, and 93.16%, respectively, consistently outperforming several state-of-the-art methods. These results validate the effectiveness and generalization capability of the proposed model across diverse change detection scenarios.</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

CDFF-Net: Innovative Feature Fusion for Improved Change Detection in Remote Sensing

  • 期刊论文

  • 发表日期：

    2025年7月17日

查看全文
  • Yunyu Liu,
  • Wangzhong Liu,
  • Jinpeng Yuan

摘要

With the rapid development of artificial intelligence and remote sensing imaging, change detection (CD) has become increasingly critical in applications such as urban planning, disaster assessment, and environmental monitoring. However, existing methods still struggle to effectively fuse bi-temporal features and accurately identify change regions due to challenges such as scale variation and semantic drift. To address these issues, this paper proposes a novel deep learning-based framework, the Change Detection Feature Fusion Network (CDFF-Net), which enhances the representational power of fused features and improves change detection performance. Built upon the U-Net backbone, CDFF-Net introduces two key modules: (1) the Change Guidance Module (CGM), which dynamically recalibrates deep features using change priors to improve sensitivity to true changes; and (2) the Parallel Partial Convolutional Attention (PPCA) module, which maintains the integrity of salient features through multiscale selective enhancement during bi-temporal fusion. Extensive experiments on three benchmark remote sensing datasets Sun Yat-sen University Change Detection Dataset (SYSU-CD), Large-scale Building Change Detection Dataset from LEVIR (LEVIR-CD), and Wuhan University Change Detection Dataset (WHU-CD) demonstrate that CDFF-Net achieves F1-scores of 81.66%, 92.14%, and 93.16%, respectively, consistently outperforming several state-of-the-art methods. These results validate the effectiveness and generalization capability of the proposed model across diverse change detection scenarios.