The WPD-EEMD Method And Its Applied Research In The EEG Signals Processing

Ensemble empirical mode decomposition(EEMD), which is aimed at non-stationary signals, was proposed by N. E. Huang. EEMD is a noise-assisted data analysis method based on empirical mode decomposition(EMD) to solve the mode mixing problem in EMD. Through processing groups of white noise-added signals by EMD, it treats the mean value as the final true result. Because of the zero mean value characteristic, those white noises will be cancelled out. The most important effect of the added white noise is to make the signal evenly distributed in time-frequency space, so the different components of signal can be mapped to the appropriate scales. But, with an increasing number of white noises, the running time is longer, especially in the condition of large decomposed signal. In order to solve this problem, this thesis put forward a new method called WPD-EEMD. Taking the EEG signal processing as background, experiment results demonstrate the validity of this method. In the process, this thesis has mainly done following work:Firstly, EEG signal is a kind of typical non-stationary signal, which contains some abrupt information. To verify EEMD whether a valid method in the EEG signal processing, a amount of experiments has been done in this thesis. The simulations prove that EEMD is effective in detecting abrupt information and improving mode mixing.Secondly, to explore how to improve the EEMD decomposition efficiency more effectively, this thesis studied the wavelet de-noising method, which is widely used in removing noises from EEG signals. It only subdivides the low frequency part of a complex signal, but ignore the high frequency part. It is easy to loss some significant abrupt information, so wavelet packet de-noising method was introduced in this thesis. It can save quick changed features at the same time of removing noises. By making experiments with an open EEG data sequence, it was verified that wavelet packet de-noising method has more superior effect than wavelet.Finally, it can achieve more satisfactory results by using EEMD method when the number of white noises is more than one hundred or even hundreds. Each time a white noise added, the EMD decomposition executed once more. So, EEMD takes a long computation time. Concerning this issue, this thesis proposes a new method called WPD-EEMD for EEG signal processing with the combination of wavelet packet and EEMD. By improving the signal-to-noise ratio, it can reduce the number of iterations of every EMD process to reduce the computation time of EEMD. Both WPD-EEMD and traditional EEMD and are used to analyze the actual EEG signals. And comparative experiments demonstrate that this WPD-EEMD tool is more effective than traditional EEMD method in EEG signal processing.