Extract EEG Features by Combining Power Spectral Density and Correntropy Spectral Density

Abstract

The electroencephalogram(EEG)-based brain computer interface (BCI) has been applied to many fields, such as medication, old-age help, transportation and entertainment. How to extract efficient features from low signal to noise ratio (SNR) EEG signals is one of the challenges in EEG signal analysis. In EEG-based BCI systems, power spectral density (PSD) is an efficient and widely used frequency feature, but the performance may degrade seriously when applied to data with low SNR. To improve the performance of EEG feature extraction, in this paper, we attempt to use the correntropy spectral density (CSD) as an EEG feature, and further combine the CSD and PSD together to construct a combinatorial EEG feature, called CSD & PSD. Firstly, an experiment on artificial EEG data is performed to validate that the CSD is more suitable for non-Gaussian and low SNR signal analysis. Then, the PSD, CSD and CSD & PSD are used to extract features of real motor imagery (MI) EEG signals. The results show that CSD & PSD often outperforms both PSD and CSD in different noise scenarios, which validate that the weighted combination feature CSD & PSD inherits the characteristics of CSD and PSD, and it is more suitable for EEG signal analysis in various SNRs.