AN ADAPTIVE NEURAL NETWORK ARCHITECTURE FOR TARGET TRACKING AND RECOGNITION

THIS PHASE I RESEARCH WILL PRODUCE A DESIGN FOR AN ADAPTIVE NEURAL NETWORK ARCHITECTURE THAT WILL PERFORM TARGET RECOGNITION AND TRACKING OF IR DATA. THIS WORK WILL DESCRIBE TWO NEURAL NETWORK MODELS THAT WILL IMPLEMENT A MOTION-BASED TRACKING PARADIGM WHICH WILL LOCATE, EXTRACT, IDENTIFY AND TRACK TARGETS FROM IR DATA. THE PARADIGM PERFORMS THESE FUNCTIONS FOR TARGETS WHICH ARE MOVING RELATIVE TO THE BACKGROUND. THE MODELS DESIGNED FOR PHASE I CONSIST OF A MOTION SEGMENTATION NETWORK FOR EXTRACTING SPATIAL PATTERNS AND A SEQUENCE RECOGNITION NETWORK FOR DETERMINING TARGET IDENTIFY BASED ON MULTIPLE SAMPLES. IN PHASE II OTHER NETWORKS WILL BE ADDED, INCLUDING A DUAL-SAMPLING METHOD FOR PROCESSING MOTION DATA AND A CONTROL NETWORK FOR COMBINING INFORMATION FROM THE DUAL SAMPLES TO PRODUCE ROBUST MOTION-BASED RECOGNITION AND TRACKING.