Study and analysis on anomaly detection methods for hyperspectral imagery

Abstract

Anomaly detection is helpful in many applications such as food monitoring, production testing, security surveillance, military countermeasure and so on. Spectral imaging technique is often resorted to for accurate abnormal target discrimination due to its high-resolution spectral/spatial information acquisition ability and a great number of data processing methods. Anomaly detection methods for hyperspectral imagery are contrastively studied in this paper. A self-developed visual-band hyperspectral imaging spectrometer is adopted to collect data cubes of certain experimental scene before two kinds of spectral-domain descriptors are used to execute abnormal camouflage detection. Detection effect of information divergence and generalized angle that are utilized as detection descriptors is visually and quantitatively compared and time consumption is assessed. The study is proved to be of significance to meet the anomaly detection demand that is based on spectral signature comparison and can be developed for further detection descriptor study and other imaging techniques beyond spectral imaging.