From Attention Stacking to Architecture Rethinking: Dual-Path Decoupling for Efficient Remote Sensing Image Segmentation
IEEE Geoscience and Remote Sensing Letters · 2026 · 23 · 1-5
DOI: 10.1109/lgrs.2026.3672712 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Geoscience and Remote Sensing Letters, 2026. Verfasst von Zhaoyao Zhang und 2 weiteren Autor:innen. Die Arbeit nennt 13 Literaturangaben. Gefördert von: First Class Discipline of Zhejiang-A.
Abstract (Originalsprache)
Remote sensing image segmentation has long struggled with balancing efficiency and accuracy. Most current research focuses on hybrid architectures that combine state-space models with convolutional neural networks (CNNs), with CNNmamba UNet (CM–UNet) being a notable example. However, its single-path decoding mechanism creates a performance bottleneck, and its multilevel homogeneous attention mechanism introduces significant computational redundancy. This letter proposes a novel dual-path U-Net (DP-UNet), using CM-UNet as the baseline. Our key insight is to explicitly decouple the processing of global context and local details, which are inherently distinct tasks. First, we innovate by designing a decoupled visual state-space (DVSS) block: one path models long-range dependencies through a simplified state-space model, while the other path integrates efficient channel attention (ECA) and a new parameter-modulated convolution (PMC) module to enhance detail extraction. The features from both paths are fused through an adaptive path fusion gate (APFG) module. Second, we streamline attention by removing redundant units from the SS2D module in CM-UNet and simplifying its multiscale fusion module into a multiscale spatial kernel (MSK) module, which refines its structure to focus solely on spatial attention, reducing model complexity. Experimental results on three benchmarks show that DP-UNet outperforms existing state-of-the-art methods in both accuracy and efficiency. This study demonstrates the potential of explicit decoupling architectures in visual tasks. The codes are available at https://github.com/sharkzzzy/DP-UNet
Förderung
- First Class Discipline of Zhejiang-A
Zitieren
APA
Zhang, Z., Sun, J., & Yang, Z. (2026). From Attention Stacking to Architecture Rethinking: Dual-Path Decoupling for Efficient Remote Sensing Image Segmentation. IEEE Geoscience and Remote Sensing Letters, 23, 1-5. https://doi.org/10.1109/lgrs.2026.3672712
Vancouver
Zhang Z, Sun J, Yang Z. From Attention Stacking to Architecture Rethinking: Dual-Path Decoupling for Efficient Remote Sensing Image Segmentation. IEEE Geoscience and Remote Sensing Letters. 2026;23:1-5. doi:10.1109/lgrs.2026.3672712
BibTeX
@article{Zhang2026,
author = {Zhang, Zhaoyao and Sun, Jie and Yang, Zhuo},
title = {From Attention Stacking to Architecture Rethinking: Dual-Path Decoupling for Efficient Remote Sensing Image Segmentation},
journal = {IEEE Geoscience and Remote Sensing Letters},
year = {2026},
volume = {23},
pages = {1--5},
doi = {10.1109/lgrs.2026.3672712}
}
EUVIMED