Feature Extraction of EEG Signal by Power Spectral Density for Motor Imagery Based BCI

Signals produced from the brain are widely known as Electroencephalogram (EEG) signal interfacing with any communication device creates a unidirectional communicating channel in the absence of neuro-muscular pathways. An effective Brain-Computer Interface (BCI) system basically consists of three operations which are signal recording, feature extraction and classification. Efficient and reliable classification of EEG signal for motor imagery (MI) based BCI system depends on the accuracy of denoising and extracted features of the signal. Extracted features are intended to be lossless key information obtained from a signal that describes a dataset accurately. It is important to minimize the classification complexity and maximize the accuracy. Traditional strategies can be used to process the signal, but the diverseness of the EEG signal conceivably could not be depicted utilizing a linear analytical approach. Hence, this paper adopted the power spectral density (PSD) feature extraction technique to extract the features based on various frequency transformations that enhance the classification performance. Graz BCI competition IV, dataset 2b has been utilized in this paper that consisting of two different classes of motor imagery left-hand and right-hand movement. Overall, 0.61 of Cohen’s Kappa accuracy obtained using the LDA classifier.