A New Way of Channel Selection in the Motor Imagery Classification for BCI Applications

A New Way of Channel Selection in the Motor Imagery Classification for BCI Applications

Nowadays, motor imagery classification in electroencephalography (EEG) based brain computer interface (BCI) systems is a very important research topic in the study of brain science. As EEG contains multi-channel EEG recordings with huge amount of data, it is sometimes very challenging to extract more representative information from original EEG data for efficient classification of motor imagery (MI) tasks. Thus, it is necessary to diminish the redundant information from the original EEG signal selecting appropriate channels and also to reduce computational cost. Addressing this problem, we intend to develop a methodology based on channel selection for classification of MI tasks in the BCI applications. In this study, we introduce a new way of channel selection considering anatomical and functional structural of the human brain and also investigate its impact in the classification performance. In this proposed method, at first we select the channels from motor cortex area, and then decompose EEG signals using wavelet energy function into several bands of real and imaginary coefficients. The relevant band’s coefficient energy has been used as feature vector in this research. After that, the extracted features are tested by three popular machine learning method: Linear Discriminant Analysis (LDA), Support Vector Machine (SVM) and K-Nearest Neighbour (KNN). The method is evaluated on a benchmark dataset IVa (BCI competition III) and the results demonstrate classification improvement with less computational cost over the existing methods.

Files associated with this item cannot be displayed due to copyright restrictions.