Pattern recognition expert system for mine classification and detection sonar

The authors present a pattern-recognition expert system for use with mine classification and detection sonar. The expert system detects and classifies mine-like targets based on the properties of the principal elastic response on the return sonar signal. Information extracted from the time delay and frequency of the principle elastic response signal is used to discriminate between mine-like objects and actual mines. The expert system extracts properties from a sliding window power spectral density on the return signal. From these extracted properties decisions are made on the characteristics of the signal within each particular time window. For each time window the expert system decides on the presence of: (i) narrowband noise, (ii) broadband noise, (iii) elastic return, and (iv) the main spectral return. The system was tested on both simulated and real scale-model data. Results on simulated data showed the successful detection of the elastic signal in both reverberation-dominant and narrowband-noise-dominant ewnvironments