Kernel subspace-based real-time anomaly detection for hyperspectral imagery

Taking full advantage of nonlinear information, kernel-based nonlinear versions of anomaly detection algorithms generally gain wide attention in hyperspectral imagery. Kernel RX algorithm is not new but a real-time procedure of KRX has not been explored in the past. The need of real-time processing arises from the fact that many targets especially moving targets, must be detected on a timely basis. This paper presents a real-time anomaly detection algorithm based on KRX, named as real-time causal kernel RX detector (RTCKRXD) by which hyperspectral image data can be processed timely. Experimental results demonstrate the new real-time version of KRX significantly solves real-time processing problem compared to conventional KRX anomaly detector with a comparable detection performance.