Detection and recognition research of underwater noise based on multi-scale statistical model in wavelet domain

The statistics modeling method of underwater noise signal is studied. The hidden Markov tree model in wavelet domain is adopted to model the sea noise and various ship-radiated noise. Using HMT models, properties of HMT model for ship-radiated noise are thoroughly studied by experiments in different work conditions and sea conditions. The results can instruct to set appropriate values for parameters of the HMT model. Moreover, using difference between models of the sea noise and that of the ship-radiated noise, a method of detecting ship-radiated noise from sea noise is put forward. Experiments proved that the method outperform methods which are based high order statistics analysis, zero-crossing detection and energy detection respectively. Furthermore, an improved classification approach based on HMT model is presented, which integrates the wavelet coefficients HMT models with support vector machine. The performance of this approach is evaluated experimentally in classifying of four types of acoustic noises. With an accuracy of more than 90%, this HMT-based approach is found to outperform previously proposed classifiers.