Motor imagery EEG based brain-computer interfacing using Fourier-Bessel series expansion

Abstract

The electroencephalography (EEG)-based brain-computer interface (BCI) is a most emerging tech nology incorporated into treating patients suffering from cognitive or physical impairments. EEG signals are the recording of brain electrical activity. One of the most well-known ways to deal with BCI is motor imagery (MI). The MI-based BCI is an intuitive interface that directly controls computer applications from brain activity. We present a Fourier-Bessel series expansion (FBSE) based classifi cation framework for MI from recorded EEG signals for enhancing the BCI application. The FBSE spectrum has a better spectral resolution for the non-stationary signals than the Fourier spectrum. It provides a representation of real signals in terms of positive frequencies, and it does not require the use of a window function in order to obtain the spectrum of the signal. Due to its unique and compact representation, The FBSE decomposition method is used for MI-specific rhythm separation from EEG signals.The proposed work aims to enhance rhythm separation and feature extraction and provide improved classification of MI-EEG motor imagery task. Then multi-domain features were extracted from EEG rhythm, namely Hjorth and band power. The Hjorth and band power features are estimated from enhanced EEG signals. Further, the obtained features were tested based on dif ferent classifier networks, To classifyof the right hand, left hand, both feet, and tongue movement of MI task. The classifier, namely linear discrimination analysis (LDA), k-nearest neighbours (k-NN), and ensemble k-NN, are employed. In comparison, classification network k-NN provides the best classification accuracy on obtained features from FBSE based rhythm.