Automatic Target Recognition Using Genetic Algorithms and Stochastically Deformable Templates

The success of DoD's thrusts in Global Surveillance and Precision Strike depends critically on the ability to perform Automatic Target Recognition (ATR) and Hostile Target Identification (HTI). The technologies of modeling, sensor data fusion and adaptive intelligent image processing are key to the success of ATR and HTI. An innovative approach to intelligent model-based image processing and target recognition is proposed using methods of Stochastically Deformable Templates Matching (SDTM) and Genetic Algorithms (GA). The overall ATR problem under low SNR conditions is formulated as a model-based multi-trajectory stochastic maximization problem. The global maxima are found using a parallel Genetic Algorithm. The SDTM-GA approach will be applied to real and simulated SAR data on targets of interest to the DDARPA. The data will be obtained from Westinghouse-Norden and through the DoD ATR Working Group. Westinghouse-Norden will support Phases I & II and commercialize the results in Phase III. Prof Ulf Grenander of Brown University, the originator and leading world authority on Pattern Theory and Stochastically Deformable Template Matching will serve as a consultant.