Classification of Sub-frequency Bands Based Two-Class Motor Imagery Using CNN

Abstract

EEG has been primarily used in both clinical and research applications. Brain-computer system (BCI) is one of the leading EEG research applications that offer special users a new means of communication. Previous studies have reported the occurrence of MI patterns in mu and beta rhythms, but that does not provide in-depth knowledge of the frequency range. This paper focuses on the classification of 2-class Motor Imagery using several frequency sub-bands in the mu and beta range. “EEG motor imagery dataset from the Physionet database,” has been used for validation purposes. Although this data includes both imagery and real movements, we have just used the imagination data. Data is collected from 109 healthy subjects, but we have only used the first 15 subjects in the study. The study aims to divide the data into multiple frequency bands to study the motor imagery classification behaviour over different frequencies. Afterward, a CNN-based deep learning model with two convolutional layers has been used to classify the left and right classes for different types of same data. The study seeks to compare the results from various sub-frequency bands.