Spatial-Frequency Unfolding Network With Mamba for Dual-Camera Compressive Hyperspectral Imaging

Coded aperture snapshot spectral imaging (CASSI) encodes a three-dimensional hyperspectral image (HSI) into a two-dimensional measurement. Dual-camera compressive hyperspectral imaging (DCCHI) augments CASSI with a panchromatic (PAN) camera, but existing methods still struggle with long-range spatial modeling and sufficient cross-modal fusion. To solve these, we propose the spatial-frequency unfolding network with Mamba (SFUN) for DCCHI, using spatial-frequency guidance within a hybrid encoder-decoder architecture. The SFUN combines Frequency-Guided Fusion (FGF) for spectrally consistent cross-modal fusion with Guide Mamba (GM) for efficient long-range spatial modeling. Specifically, the GM incorporates Row-Column Mamba (RC-Mamba) to extract multi-scale, multidirectional spatial–spectral features, together with Cross-modal RC-Mamba (CrossRC-Mamba) for gated PAN-guided conditioning, thereby enabling bidirectional interactions between the two modalities. Experiments on both simulated and real-world datasets demonstrate that the proposed method outperforms competing methods in restoring both spectral and spatial information.