A Comparison of Multivariate Outlier Detection Methods For Finding Hyperspectral Anomalies

Hyperspectral anomaly detection is a useful means for using hyperspectral imagery to locate unusual objects. Current anomaly detection methods commonly use non-robust statistical methods that may lead to inaccurate detection results. This research explores the use of different multivariate outlier detection methods for the anomaly detection problem. Theoretically, these methods are better suited than existing anomaly detection methods for finding anomalous objects in a hyperspectral image. This hypothesis is tested by applying a range of outlier detection methods to both simulated and real-world image data. Test results indicate that multivariate outlier detection can achieve superior detector performance relative to benchmark anomaly detection methods.