Spectral Feature Gating Adapter for Hyperspectral Object Tracking

Hyperspectral object tracking has garnered significant attention due to its ability to leverage rich spectral information for robust performance in challenging scenarios. However, existing methods often rely on transferring RGB-based pre-trained networks, which suffer from spectral information loss during band selection or computational inefficiency in band regrouping strategies. To address these limitations, this paper proposes a novel spectral feature gating adapter for hyperspectral object tracking (SGAT). SGAT introduces a lightweight adapter-based architecture to optimize band regrouping, enabling efficient and adaptive fusion of spectral features. The proposed Multiple Band Adapter (MBA) module extracts discriminative spectral representations from regrouped band subsets, while the Dynamic Fusion Gate (DFG) module dynamically weights and integrates these features based on their relevance to the tracking task. By jointly leveraging the MBA and DFG, SGAT effectively preserves critical spectral information while minimizing computational overhead. Extensive experiments on HOTC dataset demonstrate that SGAT achieves state-of-the-art performance, outperforming existing trackers.