Exploiting Evolutionary Algorithms for Improved Automatic Target Recognition

"The U.S. Army is conducting extensive research involving the development of innovative algorithms for detecting and classifying high-value military targets using LADAR sensors. Many of these designs employ a model-based approach in which targets,background, and target/background interactions are modeled. A central tenet of this approach is that the large number of target, sensor, and background combinations found in realistic scenarios precludes the use of automatic target recognition (ATR)algorithms that rely solely on pre-computed templates or "brute force" model matching. These approaches can be costly and the form factor impractical for a variety of military applications. As an alternative to model-based approaches, feature-based ATRsystems have also been investigated whereby target and scene features are extracted from the data for target detection and classification. The SBIR Phase I program demonstrated a feature-based ATR system that exploited evolutionary algorithm (EA)technology. The ATR system accepted a feature vector extracted from synthetic LADAR imagery and output a target classification decision with a high probability of correct classification. This Phase II effort will expand the Phase I results by computingand processing features extracted from real LADAR imagery that has been collected by the Army and Air Force using a variety of LADAR sensors. The data encompasses a diversity of target classes imaged at v