2026· IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing· Vol 19, pp. 33825-33836· 0 citations· 24 references
TL;DR
SMGNet is proposed, a lightweight spectral-Mamba-guided network that establishes a “spectral leads, spatial follows” paradigm that employs a state space model (Mamba) to efficiently capture global long-range spectral dependencies with linear complexity, while a lightweight Res-CNN preserves intact spatial topology.
Abstract
Hyperspectral image classification faces a fundamental inductive bias conflict: reconciling the physical continuity of spectral sequences with the 2-D topology of spatial structures. Existing methods either disrupt spectral continuity [convolutional neural networks (CNNs)] or destroy spatial topology through forced flattening (Transformers and pure-sequence Mamba models), while conventional dual-branch networks rely on shallow posthoc fusion that overlooks the inherent physical leader–follower relationship between spectral and spatial modalities. To address these challenges, we propose SMGNet, a lightweight spectral-Mamba-guided network that establishes a “spectral leads, spatial follows” paradigm. The spectral branch employs a state space model (Mamba) to efficiently capture global long-range spectral dependencies with linear complexity, while a lightweight Res-CNN preserves intact spatial topology. Central to the framework is the spectral-guided feature modulation module, which transforms global spectral priors into dynamic affine parameters for stagewise conditional injection throughout the spatial network, enabling deep, semantically conditioned cross-modal interaction. Under a rigorous data leakage prevention protocol, SMGNet achieves overall accuracies of 98.88%, 99.07%, and 97.83% on the Pavia University, Salinas, and QUH-Tangdaowan datasets, respectively, with only 0.33 M parameters, demonstrating exceptional accuracy–efficiency tradeoff.
A hyperspectral conditional diffusion for dense (HCD2) classification framework is proposed in this article, which reformulates HSI classification as a conditional label-map denoising process guided by hyperspectral features, enabling generative dense semantic inference.
Hong-Qiao Wang, Hai-Yan Sun, Xuan-Chen Yan et al.· IEEE Transactions on Geoscie...· 0 citations
Using the whole image as input for hyperspectral image (HSI) classification, the model needs to handle broader spatial dependencies and more diverse spectral variations. In modeling nonlocal spatial relationships, scanning based on fixed windows is prone to directional bias, leading to the loss of spatial boundary deta...
X. Chai, Bing-Bing Lei, Wei Feng et al.· IEEE Transactions on Geoscie...· 0 citations
Hyperspectral image classification (HSIC) aims to discriminate land cover categories accurately by exploiting rich spatial and spectral information. However, existing methods still suffer from insufficient collaborative modeling of spatial and spectral features, as well as inadequate fusion of local and global informat...
Mamba has recently shown great potential for hyperspectral image (HSI) classification. However, existing Mamba-based methods often disrupt the intrinsic 3-D spectral–spatial structure during sequence construction by either embedding spectral information into channels or modeling spectral and spatial sequences separatel...
Run-Min Lei, Yu-Chuan Zhou, Xiang Zhang et al.· IEEE Journal of Selected Top...· 0 citations
Hyperspectral image classification requires a compact representation that can jointly capture discriminative spectral cues, local spatial structures, and efficient inference behavior under limited annotations. Existing CNN-based methods provide effective local pattern extraction but are less capable of modeling long-ra...
Xiao-Fei Yang, Rui-Ying Zhang, Wei-Jia Cao et al.· IEEE Transactions on Geoscie...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.