Research On Feature Extraction Methods For Non-Stationary And Non-Gaussian Ship-radiated Noise

Wavelet analysis tool is effective to deal with time-varying and non-stationary signals while Higher-Order Statistics (HOS) can suppress Gaussian noise. According to the fact that ship-radiated noise is typical non-stationary and non-Gaussian, and ambient noise meets Gussian distribution, feature extraction techniques applied for ship-radiated noise are investigated based on Wavelet analysis and HOS.The main work and originality in this paper can be summarized as below:1. Studies on de-envelope using complex analytical Wavelet transform, and harmonic components extraction of envelope through 11/2 dimensional Spectrum.Results indicate this method can effectively improve the signal to noise ratio (SNR), extrude periodic ingredient and carry out filtering while the corresponding parameters can be adjusted flexibly.2. Studies on continuous and line spectrum extraction in 11/2 - dimensionalSpectrum, and extracting feature parameters in virtue of adaptive wavelet-based neural network, which can be applied to ship target classification and recognition simultaneously.3. Studies on feature analysis and extraction of ship-radiated noise adopting bicepstrum method. Computer simulation and experiment results prove that HOS method in ship-radiated noise processing has many advantages, and the bicepstrum's performance of noise- resistant is superior to conventional power cepstrum method.