RESULTS OF USING AN ARTIFICIAL NEURAL-NETWORK TO DISTINGUISH SINGLE ECHOES FROM MULTIPLE SONAR ECHOES

Empirical results illustrate the pitfalls of applying an artificial ne ural network (ANN) to classifying underwater active sonar returns. Dur ing training, a back propagation ANN classifier ''learns'' to recogniz e two classes of reflected active sonar waveforms. Waveforms in class 1 have two major sonar echoes or peaks. Waveforms in class 2 have one major echo or peak. Our results show how the classifier ''learns'' to distinguish between the two classes. Testing the ANN classifier with d ifferent waveforms having one major peak, and waveforms having two maj or peaks generated unexpected results: The number of echo peaks was no t the feature used to separate classes.