Research On Speech And Fractal Feature Extraction Of Underwater Passive Targets

Feature extraction is the key of the targets classification. On account of the sea environment complexity and the underwater acoustic channel particularity, it is a puzzle to extract a kind of target essence feature and underwater distant detection effective feature. This paper focuses on speech feature parameter extraction and fractal feature extraction of underwater targets, which includes MFCC, multi-fractal feature in time domain and frequency domain.MFCC (Mel-Frequency-Cepstrum Coefficients) is based on the human ears' non-linear frequency characteristic and perform a high recognition rate in practical speech recognition application. Theoretically, the principles of identifying the targets by sonar soldiers and the one of speech recognition by human ears are quite similar. Thus, MFCC is applied to feature extraction of ship-radiated noise. The number of filter and order of MFCC are decided by experiment. The results show that MFCC feature coefficients have higher recognition rates for underwater acoustic target.A Fractal Brownian Motion model of ship radiated noise is established. Time delay method is used to reconstruct the phase space of time series based on Takens phase space reconstruction theory. In the hyper dimensional phase space, non-linear characteristics of time series are studied.The multi-fractal feature extraction method of ship radiated noise in frequency domain is studied. Work out the singularity measure of power spectrum. Combined with the wavelet-multi-fractal analysis, General-Dimension is given. The results show that this feature is effective.