LGMM-Net: A Local–Global Encoder and Mask Mamba Decoder Network for Remote Sensing Change Detection
IEEE Transactions on Geoscience and Remote Sensing · 2026 · 64 · 1-23
DOI: 10.1109/tgrs.2026.3662322 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Geoscience and Remote Sensing, 2026. Verfasst von Chen Fang und 4 weiteren Autor:innen. Die Arbeit wurde bisher 4-mal zitiert (Crossref) und nennt 46 Literaturangaben. Gefördert von: Natural Science Foundation of Xinjiang Uygur Autonomous Region, National Natural Science Foundation of China, National Key Research and Development Program of China.
Abstract (Originalsprache)
Remote sensing (RS) change detection (CD) is critical for applications such as urban planning and disaster assessment. It enables accurate identification of changes in ground objects across bi-temporal images. With the rapid development of deep learning, hybrid convolutional neural network (CNN)–transformer approaches have significantly advanced this field. However, existing fusion networks often lack robustness. They frequently misclassify pseudo-changes as true ones. Even with hard channel-splitting strategies that jointly model local and global features, this issue remains unresolved. To address these challenges, we propose a novel local–global encoder and mask Mamba decoder network (LGMM-Net). The network employs multikernel convolutions (multiconvs), window attention, and state-space sequence modeling to achieve robust fusion of local details and global semantics. First, we introduce a dynamic channel soft-splitting strategy within the dynamic local–global parallel encoder (DLGPE). This encoder integrates multiconvs, fixed-window attention, and shifted-window attention to capture both local details and global semantics. Second, we design the pyramid mask Mamba (PMM) module with a new scanning mechanism, termed mask-guided directional selective scan (MGD-SS2D). In this module, directional convolutions define scanning paths, while feature rotations enable four-directional modeling, thus enhancing boundary representation. Extensive experiments on four public datasets demonstrate that LGMM-Net consistently outperforms state-of-the-art (SOTA) methods in both accuracy and robustness. Our source code is available on https://github.com/Jhin1215/LGMM.
Förderung
- Natural Science Foundation of Xinjiang Uygur Autonomous Region
- National Natural Science Foundation of China
- National Key Research and Development Program of China
Zitieren
APA
Fang, C., Cheng, S., Du, A., Wu, C., & Ding, Y. (2026). LGMM-Net: A Local–Global Encoder and Mask Mamba Decoder Network for Remote Sensing Change Detection. IEEE Transactions on Geoscience and Remote Sensing, 64, 1-23. https://doi.org/10.1109/tgrs.2026.3662322
Vancouver
Fang C, Cheng S, Du A, Wu C, Ding Y. LGMM-Net: A Local–Global Encoder and Mask Mamba Decoder Network for Remote Sensing Change Detection. IEEE Transactions on Geoscience and Remote Sensing. 2026;64:1-23. doi:10.1109/tgrs.2026.3662322
BibTeX
@article{Fang2026,
author = {Fang, Chen and Cheng, Shuli and Du, Anyu and Wu, Chunpeng and Ding, Yao},
title = {LGMM-Net: A Local–Global Encoder and Mask Mamba Decoder Network for Remote Sensing Change Detection},
journal = {IEEE Transactions on Geoscience and Remote Sensing},
year = {2026},
volume = {64},
pages = {1--23},
doi = {10.1109/tgrs.2026.3662322}
}
EUVIMED