Research On Feature Extraction And Pattern Classification Method Based On Time Frequency Analysis

Nonstationary nonlinear signal has become a hot research topic, which constitutes an important part of modern signal processing. The traditional signal-processing method appears to be inappropriate for processing of nonstationary nonlinear signal, because they are effective only in the time domain, or in the frequency domain processing signal, it is not able to effectively extract the nonlinear characteristics of the signal, the time-frequency analysis method is for solving this problem. Firstly this paper introduces some signal processing methods at present. And it shows that each signal processing method are targeted and limited through analyzing each signal processing method. By analyzing feature extraction, feature selection and pattern classification for nonstationary signals, the feature of time-frequency domain is extracted by wavelet packet decomposition method. then the kernel principal component analysis is used to reduce the dimension; Intelligent classification is conducted with support vector machine. This paper sloves the problem intelligent feature extraction and pattern classification of nonstationary signals which is based on time frequency analysis.The main research work of this paper is as follows:(1) For nonlinear feature extraction of nonstationary signals, Energy features are extracted by the method of wavelet packet decomposition. Firstly, this paper introduces the basic theory of wavelet and wavelet packet, and wavelet decomposition is only for low frequency decomposition but the high frequency, so sometimes it losts some important high frequency information. The wavelet packet decomposition overcomes the shortcomings of wavelet decomposition that the wavelet can't be decomposed again, and the high frequency part of the signal can be decomposed again to obtain the useful signal of high frequency. Finally this paper analyzes the feature extraction method that is based on wavelet packet.(2) Kernel principal component analysis is used to solve the problem of the main feature selection in nonstationary signals. Firstly, the feature selection methods of dimension reduction of PCA and KPCA are introduced respectively. The two kinds of dimension reduction methods are compared with the feature extraction method. With the defined performance evaluation index, Principal component analysis and kernel principal component analysis are applied to the feature dimension reduction of nonstationary signals, the simulation results show that with the same accuracy, the kernel principal component analysis is less than the principal component analysis, the kernel principal component analysis is better than the principal component analysis; When considering the same number of principal component features, accuracy is higher than that of principal component analysis. Kernel principal component analysis has a better effect than principal component features in aspect of dimension extraction.(3) Support vector machine is used to solve the nonstationary signal feature extraction and intelligent pattern classification problems. Firstly, the theory of support vector machine and the parameter analysis of kernel parameters are introduced. The feasibility of the research content is verified by experiment simulation. Finally, the parameters of the performance of the classifier are given: sensitivity, specificity and accuracy.(4) The method is applied to the nonstationary nonlinear Epilepsy EEG signal,firstly, removal of noise from EEG signals, and then wavelet packet decomposition is used to extract the energy characteristics from the wavelet packet decomposition coefficient. Kernel principal component analysis is done to complete the feature of dimension reduction and then the support vector machine is selected to classify the energy features. The experimental results show that the wavelet packet decomposition based on time frequency analysis is used to extract the energy characteristics of the nonstationary Epilepsy EEG signal, kernel principal component analysis selection the feature. Support vector machine used for intelligent recognition is a very good classification accuracy.