WAVELET AND LEARNING CLUSTERING ALGORITHMS FOR AUTOMATIC TARGET RECOGNITION

THE PROPOSED RESEARCH AND DEVELOPMENT EFFORT AIMS AT DEMONSTRATING THE FOLLOWING TECHNICLA CAPABILITIES; (A) HIGH PERFORMANCE ALGORITHMS FOR ATR BASED ON TWO-DIMENSIONAL SENSOR DATA; (B) EFFICIENT AND FAST SOFTWARE AND HARDWARE IMPLEMENTATIONS OF THESE ALGORITHMS ON LOW COST MULTIPROCESSOR BOARDS. THESE CAPABILITIES REST WITH THE FOLLOWING TECHNICAL PRINCIPLES; (A) THE RECENTLY DISCOVERED PROPERTIES OF WAVELET REPRESENTATIONS AS ROBUST MULTIRESOLUTION REPRESENTATIONS OF ONE- AND TWO-DIMENSIONAL SIGNALS (E.G., RADAR AND ACOUSTIC WAVEFORMS, IMAGES); (B) THE SUPERIOR PROPERTIES OF THE LEARNING VECTOR QUANTIZATION AND SELF ORGANIZING FEATURE MAP ALGORITHMS AS LEARNING CLASSIFIERS; (C) THE KEY IDEA OF LINKING THE FEATURE EXTRACTOR (PREPROCESSOR) AND THE CLASSIFIER VIA FEEDBACK; (D) THE AVAILABILITY OF NEURAL AND PARALLEL DIGITAL MULTI-CHIP BOARDS FOR HARDWARE IMPLEMENTATIONS. ANTICIPATED BENEFITS: - COMPRESSION OF HIGH RESOLUTION RADAR RETURNS. - GROUND VEHICLE ACOUSTIC SIGNATURE DETECTION. - EARLY FAULT AND DAMAGE DETECTION IN MACHINERY. - SENSOR FUSION. - IMPROVED OPTICAL CHARACTER RECOGNITION. - RECOGNITION OF INDUSTRIAL OBJECTS IN MANUFACTURING.