The Application Of Higher Order Statistic In Extraction Of Mine Characteristics

Autocorrelation in time domain and power spectrum in frequency domain are basic methods in feature extraction of underwater acoustic targets echoes. But the phase information in each frequency component cannot be contained in power spectral estimation, and power spectrum and autocorrelation only perfectly describe a Gaussian process which has a known mean. Higher-order Statistic (HOS) is proposed due to many limitations of 2-order statistic (power spectrum and autocorrelation). HOS is an efficient analysis tool in analyzing non-Gaussian signal and nonlinear signal. Extracting features and analyzing the non-Gaussian and nonlinear of mine echoes is the main contribution in this paper.The bispectrum, bicoherence, 1.5-dimension spectrum and 2.5-dimension spectrum are applied in this paper. Bispectrum and bicoherence are a third-order spectrum. Bispectrum describe the energy location of the signal in bi-frequencies and the non-Gaussian and the phase-coupling can be characterized by means of bicoherence. The 1.5-dimension spectrum and 2.5-dimension spectrum of complex signals can detect separately 2-order and 3-order phase coupling relations. The typical signals, such as CW pulse signal, band-limited Gaussian noise and reverberation from Unit-scatter Model, are be researched by means of power spectrum, bispectrum and bicoherence. A number of simulations are done based on 1.5-dimension spectrum and 2.5-dimension spectrum. Besides, the 2-order and 3-order nonlinear phase coupling phenomenon are researched in Gaussian noise, Exponential distribution noise and Rayleigh distribution noise environment by 1.5-dimension spectrum and 2.5-dimension spectrum, which are demonstrated not only to suppress symmetry distribution noise but also non-symmetry distribution noise.In the end of this paper, a large number of experiment data sets are tested applying bispectrum and bicoherence. Feature extraction and optimization are done based on bispectrum and bicoherence. The reverberation and the target echoes have good class separability in bicoherence feature space. But the feature based on bispectrum is proved a bad feature vector. Lastly, the different subsections of semis-buried steel mine shell echoes are analyzed by means of 1.5-dimension spectrum and 2.5-dimension spectrum. Consequently, the mine subsection echoes has different nonlinear phase coupling degree comparing other subsections echoes. These other subsections echoes are reverberations from the bottom zones around mine shell.