SENSOR DATA CORRELATION/CLASSIFICATION VIA FUZZY CLUSTERING

C-LANGUAGE SOFTWARE SHALL BE DEVELOPED THAT WILL ACCOMPLISH A CRISP CLUSTERING OF ASW CONTACT REPORTS FROM ESM, SONAR, AND OTHER SOURCES. CONTACTS SHALL BE GROUPED TOGETHER SO THAT CONTACTS IN THE SAME GROUP ORIGINATE FROM THE SAME OBJECT. THE PROBABILITY THAT EACH OBJECT IS A GIVEN TYPE SHALL BE ESTIMATED. THE PROBLEM SHALL BE PARTITIONED SO THAT POSITIONAL AND SIGNAL ATTRIBUTE DATA ARE PROCESSED SEPARATELY. POSITIONAL DATA SHALL BE PROCESSED BY A NEURAL NETWORK, WHICH PROVIDES A FUZZY CLUSTERING OF CONTACTS BASED ON AN APPROPRIATE MOTION MODEL. A FUZZY CLUSTERING OF CONTACTS ONTO THREAT TYPES CAN BE COMPUTED DIRECTLY BY WAY OF A THREAT DATA BASE. THE CONTACT AND THREAT CLUSTERS SHALL BE COMBINED TO FORM A CRISP CLUSTER OF CONTACTS. EACH CLUSTER OF REPORTS SHALL BE ASSOCIATED WITH ADDITIONAL INFORMATION ON THE PROBABLE OBJECT. CONTACT MOTION AND DISTRIBUTION MODELS SHALL BE DEVELOPED AS PART OF THE CONTRACT AND SHALL BE USED TO TRAIN THE NEURAL NETWORK AND TO DIRECT THE DESIGN OF THE FUZZY CLUSTER COMBINER. THE NEURAL NETWORK TRAINING SHALL BE ACCOMPLISHED IN TWO STAGES, FIRST WITH PRELIMINARY, AND THEN WITH REFINED CONTACT MODELS.