.png){kind=icon}
.png){kind=icon}
Return
Dual-branch attention network with multi-level spectral-spatial fusion for hyperspectral image classification
.png){kind=icon}
PRE
AI
D
D
L
DOI：10.1016/j.knosys.2026.115983.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Hyperspectral image (HSI) classification is a challenging task due to the high dimensionality of spectral data and complex correlations between spectral and spatial domains. Existing approaches generally assign uniform weights to all bands and apply shallow fusion, limiting the exploitation of discriminative information and introducing redundancy. To address these limitations, we propose a dual-branch multi-level spectral-spatial feature fusion (DMSSFF) network. The proposed model integrates Transformer and convolutional neural networks (CNNs) to jointly enhance spectral-spatial representations. A grouped cross-block interactive self-attention (GCISA) branch extracts detailed spectral features and learns representations sensitive to similarity, while a multi-layer shifted window spatial feature enhancement (SW-SFE) branch captures multi-scale spatial-spectral dependencies. An efficient depth-weighted cross-attention fusion (E-DCF) module further integrates the complementary features through adaptive attention and weighting. Experiments on three benchmark datasets demonstrate that DMSSFF achieves superior accuracy over state-of-the-art methods, confirming its ability to effectively capture and fuse spectral-spatial knowledge for robust HSI classification. The corresponding code is available at https://github.com/cgma-1/DMSSFF.
Keywords:
Grouped self-attention
Shifted-window spatial enhancement
Depth-weighted spectral-spatial fusion
CNN-transformer
Hyperspectral image classification
Journal
K
IF:
7.6
Papers:
1.2W
Citations:
4.5W
Organization
Cited Papers