FCdDNet: Feature cross-domain decoupling network for remote sensing change detection
文献类型：期刊论文
| 作者 | Wang, Bin1,2; Wang, Bo1,2; Qin, Pinle2; Zeng, Jianchao2 |
| 刊名 | PATTERN RECOGNITION
 |
| 出版日期 | 2026-09-01 |
| 卷号 | 177页码:113259 |
| 关键词 | Remote sensing change detection Disentangled representation Prototype contrastive learning Over-expectation push-pull loss |
| ISSN号 | 0031-3203 |
| DOI | 10.1016/j.patcog.2026.113259 |
| 产权排序 | 2 |
| 文献子类 | Article |
| 英文摘要 | Remote sensing change detection (RSCD) based on deep learning networks has achieved remarkable results. However, current deep neural networks for RSCD often use mixed feature extraction and fusion, which can easily lead to blurred prediction category boundaries. To address this issue, this paper proposes a RSCD model based on cross-domain feature disentanglement. First, the depth feature information of bi-temporal images is extracted through a siamese U-Net network, and the effectiveness of feature extraction is ensured through process supervision via the cross-entropy. Next, the public features of bi-temporal images and the unique features for each phase are directly obtained through a cross-domain feature separator, enabling the disentangled representation for changed and unchanged features. Meanwhile, the completeness of the disentangled features is ensured by a reconstruction loss function. Then, the prototype contrastive learning for public and unique decoupled feature is designed so that unreliable pixels can align with the prototypes calculated by reliable pixels. Finally, to better facilitate the CD pixel classification, this paper proposes an over-expectation push-pull loss regularization term, which aims to enlarge the inter-class distance by enhancing the predicted expectation to push or pull the positive and negative feature apart. Experiments have shown that the proposed method has achieved significant improvements in both qualitative and quantitative metrics. |
| URL标识 | 查看原文 |
| WOS研究方向 | Computer Science ; Engineering |
| 语种 | 英语 |
| WOS记录号 | WOS:001693442400001 |
| 出版者 | ELSEVIER SCI LTD |
| 源URL | [http://ir.igsnrr.ac.cn/handle/311030/220939]  |
| 专题 | 资源与环境信息系统国家重点实验室_外文论文 |
| 通讯作者 | Wang, Bin |
| 作者单位 | 1.Chinese Acad Sci, State Key Lab Resources & Environm Informat Syst, Inst Geog Sci & Nat Resources Res, Beijing 100101, Peoples R China 2.North Univ China, Dept Data Sci & Technol, Taiyuan 030051, Peoples R China; |
| 推荐引用方式 GB/T 7714 | Wang, Bin,Wang, Bo,Qin, Pinle,et al. FCdDNet: Feature cross-domain decoupling network for remote sensing change detection[J]. PATTERN RECOGNITION,2026,177:113259. |
| APA | Wang, Bin,Wang, Bo,Qin, Pinle,&Zeng, Jianchao.(2026).FCdDNet: Feature cross-domain decoupling network for remote sensing change detection.PATTERN RECOGNITION,177,113259. |
| MLA | Wang, Bin,et al."FCdDNet: Feature cross-domain decoupling network for remote sensing change detection".PATTERN RECOGNITION 177(2026):113259. |
入库方式： OAI收割
来源：地理科学与资源研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。