Application of the zero-crossing rate, LOFAR spectrum and wavelet to the feature extraction of passive sonar signals

In this paper, the features extraction of passive sonar signals and classification recognition of underwater target are introduced. Due to the complexity and non-stationary of underwater signals, the zero-cross ratio is first used to initially classify the noise signal; then the LOFAR spectrum reflecting non-stationary signal is extracted, and during which the wavelet transform is carried out for some classes of signals. Finally, a fuzzy ART neural network is constructed to carry out the classification. Results of the experiment show that, for six-class target 147 running environments, 5000 realistic data of ship, the mean correct ratio achieves 89%. The result obtained is satisfactory.