The goal of this research project is the development of a pattern recognition expert system for automated classification and detection of sonar signals. The expert system will be composed of two parts: I) front end signal processing system to extract power spectrum parameters, and II) an expert system which will make decisions on the power spectrum parameters based upon a rule-based library. The ultimate goal of this effort is an automated system which can detect and classify both active and passive sonar signals in noisy environments (less than -I0 dB SNR). Based upon preliminary work in this area, an expert system has been previously developed which can effectively detect narrowband and broadband signals in -20 dB to -30 dB SNR environments. Application of additional classification rules will further enhance the existing system so that unique classification capabilities can be achieved in noisy environments. The Phase I proposed research will be composed of the following stages: I) computer modeling of expert system, II) rule-base development, III) system integration and real-time testing. Once the basic principles of Phase I have been established, Phase II & iii efforts will incorporate full-scale developmental efforts towards an operational system.
