Skip to content
Open access

Snake Scan Mamba With Spatial-Enhanced Gated Attention for Hyperspectral Anomaly Detection

2026 · IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing · Vol 19, pp. 32730-32744 · 0 citations · 41 references

Abstract

Accurate background estimation is pivotal for hyperspectral anomaly detection (HAD). However, existing methods struggle to balance global receptive fields with local spatial continuity and are often susceptible to background contamination from noise or outliers. Leveraging the linear complexity of State Space Models (SSMs), this article proposes a robust framework named S$^{3}$M-AD. Specifically, the framework integrates an end-to-end Spectral Residual Denoising Module to prepurify spectral noise while strictly preserving full spatial resolution and pixel-level details. To address the spatial neighborhood fragmentation faced by standard SSMs when processing flattened 2-D sequences, we introduce an externalized Snake Scan strategy. This strategy strictly preserves local topological continuity prior to sequential modeling, enabling the Mamba backbone to capture the global spatial-spectral dependencies under stationary dynamics. Complementing this, we propose a Gated Spatial-Sparse Attention Mechanism to mitigate outlier interference in neighborhood reconstruction. By integrating a dynamic thresholding operator with a Non-Zero-Sum Gating Fusion mechanism, this module not only adaptively screens for high-confidence background samples, but also intelligently regulates the injection ratio of contextual information. By leveraging the causal history accumulation of the Mamba backbone, the proposed method establishes a significant reconstruction error margin between anomalies and the background. Comparative results across multiple datasets validate the superiority of S$^{3}$M-AD, achieving competitive AUC scores against state-of-the-art techniques and confirming the immense potential of spatially-aware SSMs in hyperspectral analysis.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.