Hyperspectral Anomaly Detection with Enhanced Spectral Graph Transformer Network

Hyperspectral imaging (HSI) plays a pivotal role across diverse sectors—including agriculture, environmental monitoring, and defense—by capturing rich spectral information that enables fine-grained material discrimination. Despite its advantages, the inherently high dimensionality and complexity of hyperspectral data present substantial challenges for reliable anomaly detection. Conventional approaches often fall short when confronted with subtle or context-dependent anomalies, underscoring the need for more advanced methodologies. In this study, we propose the Enhanced Spectral Graph Transformer Network (ESGTN), a novel framework that synergistically combines graph-based modeling, transformer architectures, and hyperbolic space embedding to improve the accuracy and efficiency of spectral anomaly detection. By representing HSIs as graphs, ESGTN effectively models both spatial and spectral relationships. The transformer component, equipped with self-attention mechanisms, adaptively emphasizes salient features, while the incorporation of hyperbolic embeddings provides a compact and distortion-minimized representation of the data’s hierarchical structure. Extensive experiments conducted on multiple benchmark hyperspectral datasets demonstrate that ESGTN consistently outperforms existing state-of-the-art methods, achieving superior precision, recall, and computational efficiency. These findings highlight the model’s robustness and practical applicability across a range of real-world scenarios. Moreover, this work contributes to the growing body of research at the intersection of deep learning and hyperspectral imaging, offering a scalable path forward for tackling the complex analytical demands inherent to high-dimensional remote sensing data.