MORPHOLOGY-BASED FRONTEND PROCESSING FOR IR-BASED ATR SYSTEMS

IT IS PROPOSED TO APPLY AN INNOVATIVE AND REVOLUTIONARY CLASS OF SIGNAL PROCESSING TECHNIQUES TO THE PROBLEM OF FRONTEND PROCESSING FOR IR-BASED ATR SYSTEMS. THESE TECHNIQUES OFFER THE POTENTIAL FOR SUBSTANTIAL REDUCTION IN COMPUTATIONAL REQUIREMENTS AND HARDWARE SIZE AND WEIGHT FOR REAL-TIME TARGET DETECTION AND CLASSIFICATION. THE PROPOSED APPROACH IS BASED ON EXTENSIVE USE OF MORPHOLOGICAL FILTERING OPERATIONS THROUGHOUT THE ATR PROCESSING CHAIN. SPECIAL-PURPOSE MORPHOLOGY-BASED PROCESSORS CAN BE IMPLEMENTED USING PARALLEL ARRAYS OF TRIVIALLY SIMPLE PROCESSING ELEMENTS, LEADING TO ENORMOUS SAVINGS OVER MORE CONVENTIONAL REAL-TIME PROCESSORS. WE PROPOSE TO ADAPT EXISTING MORPHOLOGY-BASED DETECTION AND DISCRIMINATION TECHNIQUES, DEVELOPED BY US IN OTHER DOD-SPONSORED RESEARCH PROGRAMS INVOLVING RADAR, OPTICAL AND ACOUSTIC IMAGING SENSORS, TO THE IR IMAGE DOMAIN. WE WILL DEVELOP A ROBUST MORPHOLOGYBASED IR TARGET DETECTION ALGORITHM IN WHICH TARGET SIZE AND SHAPE EXPECTATIONS ARE INCORPORATED PRIOR TO THRESHOLDING, I.E., AT THE RAW PIXEL-LEVEL. WE WILL ALSO DEVELOP IR-SPECIFIC FRACTAL SIGNATURES TO BE USED FOR DISCRIMINATING TARGETS FROM FALSE DETECTIONS. THESE SIGNATURES, WHICH HAVE PROVED TO BE HIGHLY EFFECTIVE IN OTHER DATA ENVIRONMENTS, CAN ALSO BE COMPUTED USING HARDWARE-EFFICIENT MORPHOLOGICAL FILTERING TECHNIQUES. FOR TARGET CLASSIFICATION WE WILL DEVELOP ALGORITHMS THAT EXTRACT UNIQUE STRUCTURAL AND TEXTURAL PROPERTIES OF TARGET CLASSES.