Toward fewer EEG channels and better feature extractor of non-motor imagery mental tasks classification for a wheelchair thought controller

Abstract

This paper presents a non-motor imagery tasks classification electroencephalography (EEG) based brain computer interface (BCI) for wheelchair control. It uses only two EEG channels and a better feature extractor to improve the portability and accuracy in the practical system. In addition, two different features extraction methods, power spectral density (PSD) and Hilbert Huang Transform (HHT) energy are compared to find a better method with improved classification accuracy using a Genetic Algorithm (GA) based neural network classifier. The results from five subjects show that using the original eight channels with three tasks, accuracy between 76% and 85% is achieved. With only two channels in combination with the best chosen task using a PSD feature extractor, the accuracy is reduced to between 65% and 79%. However, the HHT based method provides an improved accuracy between 70% and 84% for the classification of three discriminative tasks using two EEG channels.