Spiking Morphological Mamba for Hyperspectral Image Classification
Year: 2026
Type: article
Abstract: Hyperspectral image classification (HSIC) requires effective modeling of spatial–spectral representations while maintaining computational efficiency. Existing transformer-based methods often incur high computational complexity, whereas recent morphology-based and Mamba-based approaches typically exploit either structural priors or efficient sequence modeling in isolation. To address these limitations, this letter proposes SpikingMorphMamba (SMM). This compact hybrid framework integrates learnable multiscale morphological feature extraction, parametric leaky integrate-and-fire (LIF) spiking self-attention (SpAttn), and Mamba-based selective state-space modeling (SSSM) within a unified architecture. The morphology module enhances structural awareness by capturing object boundaries and local spatial patterns, while the parametric LIF spiking attention module refines spatial–spectral token interactions through adaptive membrane dynamics and spike-driven token refinement before long-range dependency modeling by Mamba. Experimental results on the Loukia, Houston13, and OHID1-9 datasets demonstrate the effectiveness of the proposed approach, achieving overall accuracies of 84.08%, 98.10%, and 96.32%, respectively, with only approximately 105k trainable parameters.
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $2,645
OpenAlex ID: W7172342557