ARTIFICIAL NEURAL NETWORK TARGET RECOGNITION DEMONSTRATION

INNOVATIVE, REAL-TIME, NEURAL NETWORK HARDWARE TECHNIQUES FOR IMPLEMENTING DR. JOHNSON'S DYNAMIC OBJECT EXTRACTION PREPROCESSOR ALGORITHM FOR AUTOMATIC TARGET RECOGNITION WILL BE INVESTIGATED, DEVELOPED AND DEMONSTRATED. SCALABILITY TO FULL REAL-TIME 2-DIMENSIONAL IMAGERY WILL BE SHOWN. THE FEASIBILITY OF DEVELOPING A PHASE II HARDWARE PROTOTYPE FOR ALGORITHM VALIDATION AND PERFORMANCE MEASUREMENT, AND OF DEVELOPING A PHASE III OPERATIONAL SYSTEM WILL BE ESTIMATED. THIS PROBLEM WILL BE ADDRESSED BY INNOVATIVE APPLICATION OF NEW DEVICES PROVIDING ON CHIP PARALLELISM. FOUR TECHNICAL OBJECTIVES SHOULD ADEQUATELY DEMONSTRATE THE FEASIBILITY OF PERFORMING ATR PREPROCESSING WITH HARDWARE IMPLEMENTATION OF THIS ALGORITHM: 1. CANDIDATE DEVICES AVAILABLE NOW OR IN THE NEAR FUTURE WILL BE EVALUATED FOR SUITABILITY; 2. THE MOST SUITABLE WILL BE USED AS BASIS FOR THE DETAILED BLOCK DIAGRAM OF A CONCEPTUAL DESIGN, PERMITTING GENERAL CONNECTIVITY WITHIN A NEIGHBORHOOD OF THE CELL OF INTEREST IN NEURAL NET FASHION; 3. THE DESIGN WILL BE SIMULATED IN SOFTWARE; 4. THE ANALYSIS AND SIMULATION RESULTS WILL BE FED BACK TO IMPROVE THE DESIGN. ANTICIPATED BENEFITS/POTENTIAL COMMERCIAL APPLICATIONS - ATR IS RECOGNIZED BY DOD AS A CRITICAL NEED, PARTICULARLY IN THE DEVELOPMENT OF "BRILLIANT WEAPONS". POTENTIAL CIVILIAN APPLICATIONS INCLUDE LAW ENFORCEMENT, COLLISION AND TRAFFIC MANAGEMENT, FACTORY AUTOMATION AND ENABLING TECHNOLOGY FOR THE HANDICAPPED. EXPERTISE WILL BE DEVELOPED IN THE DESIGN, PROGRAMMING, INTERFACING AND SCALING OF MASSIVELY PARALLEL PROCESSORS AND NEURAL NETWORKS. KEY WORDS - AUTOMATIC TARGET RECOGNITION, ATR, OBJECT EXTRACTION,