Using QR Factorization for Real-Time Anomaly Detection in Hyperspectral Images.

摘要

Anomaly detection has been used successfully on hyperspectral images for over a decade. However, there is an ever increasing need for real-time anomaly detectors. Historically, anomaly detection methods have focused on analysis after the entire image has been collected. As useful as post- collection anomaly detection is, there is a great advantage to detecting an anomaly as it is being collected. This research is focused on speeding up the process of detection for a pre-existing method, Linear RX, which is a variation on the traditional Reed-Xiaoli detector. By speeding up the process of detection, it is possible to create a real-time anomaly detector. The window covariance matrix is our main area focus for speed improvement. Several methods were investigated, including QR factorization and tracking the change in the window covariance matrix as it moves through the image. Finally, performance comparisons are made to the original Linear RX detector.