Efficiently Computing and/or Compensating for Object Variability for Automatic Target Recognition (ATR) Applications

Target variability presents a significant problem for automatic target recognition (ATR) systems. For any type of realistic ATR, the computational burden is simply too expensive if the variability problem is ignored. For example, with a template based classifier, one must store every possible configuration of the target in order to obtain a sufficient match. As part of this proposal, GCS Systems will attempt to demonstrate the feasibility of algorithms which will map various target configurations into an invariant feature space for further discrimination or classification. Specifically, GCS will examine algorithms which will account for different target articulations, components, or deformations for objects of the same class while maintaining the discrimination capabilities for objects of different classes. GCS will employ a high fidelity electromagnetic model appropriate to the scenario considered and will study certain real world effects such as gaps, cracks, and deformations. In addition, we will examine application of the developed algorithms to real world collectors such as a synthetic aperture radar (SAR) system.