TARGET RECOGNITION USING SPECTRAL/SPATIAL TECHNIQUES

THE EMERGENCE OF EO MULTISPECTRAL IMAGING TECHNOLGOY MOTIVATES EXLORATION OF POTENTIAL IMPROVEMENTS IN AUTOMATIC TARGET RECOGNITION (ATR) SYSTEM PERFORMANCE USING NOVEL SPECTRAL/SPATIAL TECHNIQUES. SUCHCONCEPTS OFFER A PROMISING COMBINATION OF COMPLEMENTARY TECHNIQUES, WITH THE SPECTRAL COMPONENET REDUCING CLUTTER AMBIGUITY FOR SPATIAL PROCESSING WHICH SIFTS HIGH-VALUE TARGETS FROM THOSE OF NON-INTEREST, E.G., TANK FROM TRUCK. THE SPECTRAL/SPATIAL ALGORITHM CONCEPTS PROPOSED BY AERODYNE INTEGRATE AND EXPLOIT MODEL-PREDICTED KNOWLEDGE USING MARKOV RANDOM FIELDS TO REPRESENT SPATIAL DEPENDENCIES OF PRODICTED TARGET SPECTRAL ATTRIBUTES. OUR CONCEIVED MODEL-BASED SPECTRAL DISCRIMINATION PROCESS, RESISTANT TO TARGET OCCLUSIONS (E.G., CAMOUFLAGE), PRODUCES APRE-CLASSIFIED SCENE. AMBIGUITIES WITH IN THIS PROCESSED SCENE ARE SIGNIFICANTLY REDUCED UTILIZING A DYNAMIC PROCESS WHICH EXPLOITS JOINTSPECTRAL AND SPATIAL CONTEXTUAL INFORMATION, RESTORING WEAK AND TOTALLY OCCLUDED TARGET PIXELS. FINALLY POSE AND SCALE-INVARIANT RECOGNITION IS ACHIEVED USING A MODEL-BASED CONTENT-ADDRESSABLE MEMORY. THE RESULTING ALOGRITHMIC STRUCTURE IS INHERENTLY PARALLEL, PROMISING COMPUTATIONAL FEASIBILITY VIA IMPLEMENTATION ON MASSIVLEY PARALLEL NEURAL NETWORKS. THE PHASE I INVESTIGATION WILL ASSESS THE EXTENT OF EXPLOITABLE SPECTRAL DIFFERENCES BETWEEN TARGETS OF INTEREST AND CLUTTER. A SURVEY OF EMERGING MULTISPECTRAL IMAGING CAPABILITY WILL BE COUPLED WITH FOCUSED ANALYSIS AND TESTING OF KEY DETAILS OF THE CONCEPT ALGORITHMS TO DETERMINE THE FEASIBILITY OF PHASE II ALOGRITHM IMPLEMENTATION.