NEURAL NETWORKS FOR TORPEDO DETECTION

SONAR OPERATORS CAN EASILY TELL WHEN A TORPEDO SIGNATURE APPEARS BY USING THE AVAILABLE VISUAL CHARACTERISTICS. DESPITE THE FACT THAT TORPEDO ACOUSTIC SIGNATURES HAVE VERY SPECIFIC CHARACTERISTICS, IT HAS BEEN DIFFICULT TO DEVELOP AN AUTOMATED TORPEDO DETECTION SYSTEM WITH A HIGH PROBABILITY OF DETECTION AND AT THE SAME TIME A LOW PROBABILITY OF FALSE ALARM. NEURAL NETWORKS PROVIDE AN APPROPRIATE ARCHITECTURE FOR BUILDING SUCH A SYSTEM. BY DEVELOPING A SYSTEM THAT CAN EXTRACT THESE VISUAL FEATURES, SCT BELIEVES THAT IT IS POSSIBLE TO PERFORM TORPEDO DETECTION WITH GREATER RELIABILITY AND ACCURACY. THE BASIS OF NEURAL NETS FOR TORPEDO DETECTIONN (NN-TD) IS TO EXTRACT THE RIGHT FEATURES FROM THE ACOUSTIC DATA AND APPLY THE PROPER LOGICAL COMBINATIONS ON THOSE FEATURES TO PERFORM ACCURATE SIGNATURE INTERPRETATION. THE STEPS NECESSARY ARE TO REVIEW AND SELECT A NEURAL NETWORK DEVELOPMENT ENVIRONMENT, DEFINE TORPEDO SIGNATURE CHARACTERISTICS, SPECIFY THE FEATURES OF SPECTRAL DATA WHICH WILL BE USED TO SUPPORT TORPEDO DETECTION, DEVELOP A NEURAL NETWORK TO IDENTIFY AND QUANTIFY THESE FEATURES, AND DEMONSTRATE THE CAPABILITY TO PRIVIDE A CLEAR CUT DECISION ABOUT THE PRESENCE OF A TORPEDO IN PASSIVE ACOUSTIC DATA.