Recognition and Feature Extraction of Motor Imagery EEG Signals Based on Complementary Ensemble Empirical Mode Decomposition

Abstract

In view of the characteristics of high nonlinearity and fractional stationarity of motor imagery EEG signals, an improved Complementary Ensemble Empirical Mode Decomposition (CEEMD) method is proposed in this paper. The method does not need to select the basis function in advance and has the characteristics of high self-adaptability. In this paper, Hilbert transform, mutual information, sensitive factor and approximate entropy are used to obtain the time-frequency characteristics, recognition accuracy and other parameters of motor imagery EEG signals in time-frequency domain, and compared with other methods. Using mutual information and sensitive factors, the IMF component with useful information of the original signals can be effectively identified by CEEMD decomposition, and the selected IMF component can be reconstructed and identified by common space model and approximate entropy. The results show that the recognition rate of EEG signal by using the combination of approximate entropy and time-frequency feature is better than that by using time-frequency feature vector alone. Compared with other algorithms, the proposed algorithm has the highest classification accuracy of 84.1%, among 80.9% for ANN algorithm, 79.6% for WT algorithm and 75.8% for EMD-HT algorithm. It indicates that the method in this paper can distinguish the motor imagery tasks, and proves its effectiveness and superiority in the extraction and classification of motor EEG signals.