Empirical Mode Decomposition And Its Application In Water Acoustics Signal Processing

Underwater acoustic signal processing is an important item in the field of information, and the underwater target signal is a kind of underwater acoustic signal. Noise reduction, feature extraction and classification of underwater target signal have very important theoretical significance and practical value for detection, tracking and recognition of underwater acoustic signal.Empirical mode decomposition(EMD) has been developed in recent years as a new data analysis method. It has been proved highly suitable for the application of time frequency analysis for the nonlinear and non-stationary processes. Some existing problems and related development trend are pointed out in this paper. To solve these problems and apply the EMD to the noise reduction, feature extraction and classification of underwater target signal, some new solutions are put forward. At present, some useful results have been obtained. The main contents and innovations are as follows:Some existing problems in the EMD and ensemble EMD(EEMD) method are analyzed. One of the major drawbacks of the EMD is the frequent appearance of mode mixing. It could not only cause serious aliasing in the time-frequency distribution, but also make the physical meaning of individual intrinsic mode function(IMF) unclear. The EEMD is an important improvement of the EMD. The new approach consists of sifting an ensemble of white noise-added data and treats the mean as the final true result. The EEMD can effectively alleviate the mode mixing phenomenon of the EMD, but the ensemble average in the EEMD also result in some new problems, such as computational cost is increased, different realizations of signal plus noise may produce different number of modes, and the reconstructed signal includes residual noise which can not be eliminated by the ensemble average.To solve above problems of the EMD and the EEMD, a new EEMD with adaptive noise(EEMDAN) is proposed. The new method can not only effectively avoid the mode mixing phenomenon, but also provide an exact reconstruction of the original signal, with a lower computational cost. The decomposition performance of the EEMDAN is superior to the EMD and the EEMD method. In addition to, a noise reduction method of chaotic signal based on the EEMDAN is discussed, and the IMF correlation coefficient method based on energy density and the average cycle is proposed and used in adaptive choice of the reconstructed IMF. The proposed method is applied to noise reduction of the Lorenz chaotic signal with different signal-to-noise ratio(SNR) and four kinds of ship radiated noise signals. Some satisfactory results are obtained.Based on the bivariate EMD(BEMD) and the statistical properties of white noise by the EMD, here the noise-assisted analysis idea is applied to the BEMD, and an algorithm of noise-assisted BEMD(NABEMD) is proposed. The influence of added white noise on the decomposition results is analyzed and discussed. In the NABEMD method, the original one-dimensional signal and added white noise are respectively used as real and imaginary part, the 2D complex data that is constructed can be processed by the BEMD, and then a noise reduction method of chaotic signal based on the NABEMD is proposed. The new noise reduction method is used in the Lorenz chaotic signal with different SNR and four kinds of ship radiated noise signals. The time domain waveform and the phase space attractor trajectories are given before and after noise reduction. Some characteristic parameters such as the noise intensity, the Lyapunov exponent, the correlation dimension and the Kolmogorov entropy are calculated before and after noise reduction. The results show that the proposed method is feasible and effective for the noise reduction of chaotic signal.Comparing with the EMD and the EEMD method, the decomposition performance of two proposed methods has been significantly improved, but the NABEMD has the smallest computational cost. Considering the performance and running time, the NABEMD is selected as the right processing method of underwater chaotic signal. To verify the feasibility that the NABEMD method is used in noise reduction, feature extraction and classification of underwater chaotic signal, the NABEMD is tried to apply to local projective algorithm(LP) which is proposed based on nonlinear theory. In the LP algorithm, the neighborhood is difficult to correctly select, and the neighborhood size has an important effect on the result of noise reduction. To overcome the problem, an improved LP method based on the NABEMD is proposed. Lorenz time series and four kinds of ship radiated noise signals are used to illustrate the improved method. The result proves that the improved method can reduce the noise more efficiently than conventional LP method.Instead of the EMD method in the Hilbert-Huang transform(HHT), an improved HHT method is proposed by using the NABEMD and applied to feature extraction and classification of underwater target signal. Based on the Hilbert spectrums of ship radiated noise, some feature parameters about IMF are given. These features include(i) the center frequency of the strongest IMF,(ii) the energy difference between the high and low frequency,(iii) the instantaneous energy variation range, and(iv) the energy entropy of the IMF. Three typical chaotic systems are chosen for experiment. They include Lorenz system, Rossler system and Henon system. The above feature parameters of these chaotic systems are extracted respectively. Simulation and experimental results show that the feature values have significant differences between different types of chaotic signal, so these feature parameters about IMF can be used as the basis for classification of different chaotic signal. And then four different types of measured ship radiated noise are chosen as sample data and tested for feature extraction and classification. The results show that(i) the center frequency of the strongest IMF, the energy difference between the high and low frequency and the energy entropy of the IMF are very effective in distinguishing four types of ship target,(ii) the instantaneous energy variation range for first types of sample is significantly different from the other three types of targets.The center frequency of the strongest IMF, the energy difference between the high and low frequency and the energy entropy of the IMF are chosen to form feature vector, underwater target classifier based on support vector machine(SVM) is designed. Selecting 30 samples as training samples and 20 samples as testing samples from each type of ship signal, the classification experiments based on SVM are conducted. The results show that the proposed method is effective for the extraction and classification of underwater target signal. It will provide important references for the detection, automatic identification and classification of underwater targets.