Spatial-Spectral Mamba for Hyperspectral Image Denoising

The hyperspectral image denoising methods based on Transformer have limitations in terms of computational efficiency and model complexity. Therefore, this paper proposes a Spatial-Spectral Mamba (SSMamba) hyperspectral image denoising model based on Mamba. This model can effectively capture the spatial features of hyperspectral images within linear time complexity. Moreover, based on the traditional channel attention mechanism, an Adaptive Low-Rank Spectral Attention Module is proposed. This module utilizes the extracted spatial features, combines the low-rank characteristics of hyperspectral data and global spectral correlations, and models hyperspectral features in the spectral dimension to extract spectral features. Through comparative experiments and visualization analysis, the effectiveness of the proposed method is verified. In addition, the analysis of the number of model parameters and complexity also confirms the lightweight advantages of this method.