DECOMPOSED INFRARED (IR) MODELING FOR MODEL BASED VISION (MBV)

MODEL BASED VISION IS A PROMISING TECHNIQUE FOR RADAR AUTOMATIC TARGET RECOGNITION. ITS APPLICATION TO INFRARED IMAGERY HAS BEEN STYMIED BY TWO PROBLEMS. FIRST, MOST IR SIGNATURE MODELS WERE DEVELOPED FOR REDESIGN APPLICATIONS AND ARE NOT WELL SUITED FOR USE IN MODEL BASED VISION. SECOND, THERE HAS BEEN INSUFFICIENT OVERLAP BETWEEN THE AI MBV COMMUNITY AND THE IR MODELING COMMUNITY. DCS PROPOSES TO COMBINE THEIR EXPERIENCE IN IR MODELING FOR ATR WITH THE AI MBV EXPERIENCE OF ADS TO DETERMINE SPECIFIC WAYS TO INCORPORATE IR MODELS INTO THE SAR EXPERT MBV TEST BED SYSTEM. TO CREATE ACCURATE, EXPLICIT REPRESENTATIONS OF PHENOMENOLOGY AND OBJECTS SUITABLE FOR IMPLEMENTATION IN MBV, DCS WILL APPLY VALIDATED LUMPED PARAMETER THERMAL MODELING TECHNIQUES WHICH AVOID PROBLEMS ASSOCIATED WITH NODAL NETWORKS. BY CREATING AN INTERMEDIATE FEATURE SPACE AND MAKING SUPPORTABLE SIMPLIFYING ASSUMPTIONS, DECOMPOSITION OF TARGET AND BACKGROUND FEATURES WILL BE ACCOMPLISHED WHICH ALLOS REASONING AT MULTIPLE LEVELS OF GEOMETRIC AND SIGNAL LEVEL RESOLUTION, EXPLICITE UNCERTAINTY ESTIMATION (BASED UPON WEATHER CONDITIONS), AGGREGATE TARGET FEATURE PREDICTION, AND AUTOMATED HYPOTHESIS GENERATION AND RESOLUTION. ADS WILL DETERMINE THE CHANGES NECESSARY TO THE SAR EXPERT SYSTEM TO ACCOMMODATE THE THERMAL MODEL AND WILL PRODUCE AN IMPLEMENTATION PLAN.