Improved Hyperspectral Anomaly Detection via High Dimensional Model Representation and Connected Component Filtering

Hyperspectral imaging provides rich spectral information by capturing many narrow bands across the electromagnetic spectrum. While the high dimensionality of such data offers detailed spectral signatures, it also demands effective modeling strategies for accurate anomaly detection. We propose a three-step algorithm based on High Dimensional Model Representation (HDMR), a decomposition technique that provides feature selection and extraction. By leveraging the structure of HDMR, we enhance the visibility of anomaly pixels of hyperspectral images and improve the detection performance. Our method is evaluated against standard algorithms such as RX, LRX, CRD, and DWEST. Results show that it improves anomaly detection accuracy and reduces false alarms, demonstrating the effectiveness of HDMR-based anomaly detection in isolating anomalous patterns within high-dimensional data.