AUTOMATED FEATURE EXTRACTION AND RECOGNITION OF NATURAL AND MAN-MADE FEATURES

LNK IS PROPOSING TO DEVELOP ALGORITHMS AND SOFTWARE FOR AUTOMATIC FEATURE EXTRACTION OF MAN MADE AND NATURAL FEATURES USING A VARIETY OF IMAGE SENSORS INCLUDING INFRARED, VISUAL AND RADAR. THE PROPOSED SYSTEM WOULD BE USED FOR BOTH GEOGRAPHIC DATABASE GENERATION AND AUTOMATIC UPDATING OF GEOGRAPHIC DATABASES FROM STRIKE AIRCRAFT. WE ARE PROPOSING AN APPROACH COMBINING NEURAL NETWORKS AND EXPERT SYSTEMS. THE PROPOSED SYSTEM IS A HYBRID NEURAL NETWORK COMBINING SEVERAL EXISTING PARADIGMS. THIS APPROACH ENABLES US TO TAKE ADVANTAGE OF THE STRENGTHS OF THE VARIOUS PARADIGMS BY APPLYING THEM TO PARTS OF THE RECOGNITION PROBLEM FOR WHICH THEY ARE MOST APPROPRIATE. THE SYSTEM REQUIRES AN INTERACTIVE TRAINING PHASE FOR INITIAL DATABASE GENERATION. DATABASE UPDATING FROM ON BOARD STRIKE AIRCRAFT IS DONE IN A FULLY AUTOMATIC MODE. NEURAL NETWORKS OFFER PROMISE OF PROVIDING THE SPEED AND ROBUTNESS REQUIRED FOR REAL-TIME UPDATING OF GEOGRAPHIC DATABASES.