Phase Space Reconstruction Based Multi-Task Classification for Motor Imagery EEG

Abstract

Electroencephalogram (EEG) is a non-stationary random signal. The commonly used feature extraction methods focus on analyzing the smoothness of EEG signals. This paper proposes a method for extracting motor imagery EEG features in phase space. The phase space reconstruction method in nonlinear time series is applying for reconstructing the EEG sequence into high-dimensional phase space. While retaining the continuity of the original signal, many nonlinear dynamic characteristics in the EEG signal can also be found. This paper analyzes the motor imagery EEG from BCI competition database, reconstructs the original one-dimensional EEG signal into high-dimensional phase space by phase space reconstruction, and the “OVR-OVO” common spatial pattern double-layer filter is used to extract the phase space CSP feature (PSCSP). Finally, the classification result on the support vector machine is obtained by voting method. The final classification results show that the classification accuracy of the PSCSP introduced in this paper is 6.39% higher than that of the competition winner. According to the Kappa coefficient evaluation standard provided by the competition, the Kappa coefficient of this method is well performed. It is 9% higher than the first winner of the competition.