TransGCF: A Unified Spatial–Spectral–Frequency Framework for Robust Hyperspectral Anomaly Detection

Hyperspectral anomaly detection (HAD) is essential for applications like environmental monitoring, surveillance, and medical diagnostics. However, prevalent reconstruction-based methods often struggle with high false alarm rates, primarily because they inadvertently reconstruct high-frequency anomaly signatures or fail to adequately model complex background variations. This limitation arises from an inability to concurrently address the distinct spatial, spectral, and frequency characteristics inherent in hyperspectral data. To overcome these challenges, we propose TransGCF, aunified spatial-spectral-frequency frameworkfor robust HAD. TransGCF explicitly decouples feature extraction into three specialized branches: a Local Transformer captures fine-grainedspatialcontext, a Graph Convolutional Network (GCN) models non-localspectralcorrelations, and a novel High-Frequency Elimination Block (HFEB) actively suppressesfrequency-domain components associated with anomalies. These complementary features are integrated via a hierarchical gated fusion mechanism, yielding discriminative residual maps with enhanced anomaly salience and superior background suppression. Comprehensive experiments on six remote sensing and two medical hyperspectral datasets demonstrate that TransGCF consistently achieves state-of-the-art detection accuracy, effectively reduces false alarms, and exhibits strong cross-domain generalization. The results validate TransGCF as a robust and efficient framework for diverse real-world HAD applications. Our code is available athttps://github.com/JzXu123/TransGCF.