HIERARCHICAL SPATIAL SPECTRAL MAMBA FOR HYPERSPECTRAL IMAGE CLASSIFICATION

Recently, Transformer-based methods have exhibited remarkable success in hyperspectral image (HSI) classification tasks, primarily due to their exceptional capability in capturing long-range dependencies. However, the self-attention mechanism inherent in Transformers necessitates quadratic computational complexity with respect to sequence length, resulting in substantial computational cost. To address this limitation, a novel end-to-end hierarchical spatial-spectral Mamba (HSSM) network is proposed for HSI classification in this paper. HSSM is based on the selective State Space Model (SSM), which excels in modeling long-range dependencies within sequential data. By integrating the hierarchical structure, HSSM comprehensively explores multi-level information, enabling precise extraction of discriminative features in HSI. Compared with existing Transformer-based methods, HSSM is capable of achieving linear computational complexity, offering superior efficiency in processing lengthy sequences.