Automatic Target Recognition Within Anomalous Regions of Interest in Hyperspectral Images

This paper proposes an automatic target recognition (ATR) methodology for airborne hyperspectral imagery in which regions of interest (ROIs) containing anomalous objects are inspected for recognition of specific targets. ROI-by-ROI processing is carried out in a fully automated fashion and does not need operator intervention, thus being suitable for in-flight applications. The ROI-based ATR methodology is developed within a multiple hypotheses testing framework, and its key strengths are in the use of a bank of flexible and robust nonparametric detectors combined with an automated method for null hypothesis discrimination and an effective decision support system. Experimental results over multiple real hyperspectral images show the effectiveness of the proposed methodology for automatic recognition of several different targets embedded in various kinds of background.