Abstract

Info

Title:: Classification of multi-class motor imagery EEG data based on spatial frequency and time-series information

Author(s):: ZHOU Jie; YANG Guoyu; XU Tao; Institute of Intelligent Control and Robotics, Hangzhou Dianzi University, Hangzhou 310018, China

Keywords:: ; Keywords: electroencephalogram signal; motor imagery; feature extraction; common spatial pattern; discrete wavelet transform; long short-term memory

Abstract:: Abstract: By combining common spatial pattern (CSP), discrete wavelet transform and long short-term memory (LSTM), a multi-class motor imagery feature extraction method based on spatial frequency and time-series information is proposed. Firstly, time-series electroencephalogram (EEG) data are obtained using a sliding rectangular window, and discrete wavelet transform is used to extract the sub-band wavelet coefficients associated with motor imagery from each EEG segment. Secondly, the further feature extraction of the wavelet coefficients is achieved by one-versus-the-rest CSP, and the obtained featuresare used as the input of LSTM. Then, the time-series output of LSTM network is averaged over time steps, and finally, Softmax classifier is used for classification. Experimental results show that the accuracy rate of the proposed algorithm is 92.23%. Compared with CSP features and CSP features with frequency or time-series information, the proposed algorithm achieves a much higher classification accuracy rate, which reveals the complementarity and effectiveness of space, frequency and time-series information.