Research On Brain Signals Based On EMD

Empirical Mode Decomposition(EMD), a novel adaptive decomposition method, which is especially suitable for nonlinear and non-stationary signal. The EMD research in electroencephalograph(EEG) field is mostly confined to one-dimensional signal processing. In order to obtain more intuitive and clearer expression and expand the research content, this thesis promotes the EEG research to two-dimensional and three-dimensional image processing field according to the characteristics of EEG(such as rhythm, concentration of frequencies, complexity and diversity of brain noise and so on).Firstly, based on one-dimensional EEMD, this paper proposes the improved EEMD algorithm adapted to the characteristics of EEG signal. The improved algorithm adaptively estimates brain feature signal from the treated raw EEG signal by signal estimation algorithm based on IMF energy and segmentation threshold. Combined with features of Gauss white noise, the improved EEMD generates the new EEG-noise which is added to EEMD decomposition by replacing the traditional Gauss white noise. The algorithm is more suitable for the specific brain signal application and can better solve the aliasing problem existed in EMD decomposition of EEG signal.Secondly, this paper innovatively introduces 2D-EEMD to EEG research field, promotes the EEG research to two-dimensional space, and makes some improvements in the two-dimensional recombinant method of 2D-EEMD in the light of EEG signals’ special features. Then, due to the shortage(pseudo-approaching of adjacent electrode data) existed in 2D-EEMD processing of EEG signal, a 3D-EEMD algorithm based on EEG is proposed, which is a preliminary exploration in multi-dimensional EMD processing field of EEG field. The experimental results show that, the improved 2D-EEMD algorithm can obtain more intuitive and clearer 2D decomposed results, and the accuracy rate of SSVEP frequency increases 16%. Compared to 2D-EEMD, 3D-EEMD decomposition pays more attention to the intrinsic details of data, the accuracy rate of feature extraction, classification and recognition is higher.Finally, based on these above-mentioned improved algorithms, combined with nonlinear theory, this thesis respectively studies the classification of the two kinds of data from motor imagery and MEG mental patients.