An Approach to Target Classification by Computer in Advanced Active Sonar Systems.

摘要

Traditionally, and to a large extent today, the responsibility for making a classification decision on a detected target rests with the sonar operator. It is his task to review all target-related information received through the various inputs available to him. From these sources he must extract only the information he considers pertinent and then correlate this information before arriving at a classification decision. The need is great for the generation of automated techniques to assist the sonar operator in selecting, extracting and correlating pertinent data. This need will become more acute in the complex advanced sonar systems which are currently under development. The present investigation included four kinds of overlapping activities: (1) a search of current literature for applicable techniques, (2) development of new analytical techniques to augment existing ones, (3) generation of computer programs to implement and evaluate these techniques, and (4) development of a design for a complete semi-automatic classification subsystem. The general classification problem is formidable, and despite the fact that the approach in this investigation was restricted to methods realizable in a real-time system, the results are necessarily partial and the recommendations tentative. Nevertheless, essential ground has been covered, and it is believed the observations made here will remain valid until the next major theoretical advances in classification techniques are achieved.