Collaborative Representation Based on the Constraint of Spectral Characteristics for Hyperspectral Anomaly Detection

Abstract

Anomaly detectors based on representation do not need to make a specific assumption about the statistical distribution of background, which will have better detection performance in anomaly detection. Generally, hyperspectral anomaly detection algorithms of collaborative representation often assume the pixels in the dual-window as the background pixels. Moreover, the constraint of sum-to-one is applied to representation weights for physical meaning. However, in the process of representation, the spectral characteristics of images are not fully utilized. In this article, the spectral angle mapping is applied to adjust the representation weights of the neigh-boring pixels around the tested pixel. Experimental results demonstrated that the proposed collaborative representation based detector using spectral characteristics to adjust weight can highlight anomalies effectively and achieve better detection performance.