Abstract
Effectively discerning spatial-spectral dependencies in HSI denoising iscrucial, but prevailing methods using convolution or transformers still facecomputational efficiency limitations. Recently, the emerging Selective StateSpace Model(Mamba) has risen with its nearly linear computational complexity inprocessing natural language sequences, which inspired us to explore itspotential in handling long spectral sequences. In this paper, we proposeHSIDMamba(HSDM), tailored to exploit the linear complexity for effectivelycapturing spatial-spectral dependencies in HSI denoising. In particular, HSDMcomprises multiple Hyperspectral Continuous Scan Blocks, incorporatingBCSM(Bidirectional Continuous Scanning Mechanism), scale residual, and spectralattention mechanisms to enhance the capture of long-range and localspatial-spectral information. BCSM strengthens spatial-spectral interactions bylinking forward and backward scans and enhancing information from eightdirections through SSM, significantly enhancing the perceptual capability ofHSDM and improving denoising performance more effectively. Extensiveevaluations against HSI denoising benchmarks validate the superior performanceof HSDM, achieving state-of-the-art results in performance and surpassing theefficiency of the latest transformer architectures by $30\%$.