Manifold learning methods for wide-angle SAR ATR

The automatic recognition and characterization of civilian vehicles in urban setting is motivated by an increasingly difficult class of surveillance and security challenges. These new ATR (Automatic Target Recognition) problems are motivated by new data collection capabilities, in which airborne synthetic aperture radar (SAR) systems are able to interrogate a scene, such as a city, persistently and over a large range of aspect angles. Learning and exploiting the additional information provided by wide-aspect signatures is key to developing successful algorithms. In this paper, we study manifold learning methods to learn informative projections of the feature space for ATR algorithm design, which is also amenable to performance prediction and analysis.