HTDC: Hyperbolic Transformer With Dual-Momentum Contrastive Learning for Hyperspectral Anomaly Detection (opens in new tab)

Hyperspectral anomaly detection (HAD) remains a challenging task due to the complex spectral variability and spatial heterogeneity of real-world scenes. Reconstruction-based methods have been widely explored for deep learning-based HAD, but they rely on the assumption that background pixels can be faithfully reconstructed while anomalous pixels cannot. Although intuitive, this dichotomy oversimplifies the spectral–spatial characteristics of hyperspectral data and often fails in practice. In p...