Classification of vision perception using EEG signals for brain computer interface
Abstract
Patients suffering from Motor Neuron Disease (MND) and semi-paralysis have
trouble to maneuver a conventional wheelchair independently. As a response, this
research was conducted whereby an individual’s visual perception can associate to
movement controls. The designed system could later on be integrated into an autonomous
wheelchair. The Brain Computer Interface (BCI) system would require the
Electroencephalography (EEG) signal to be recorded from the subject using Mindset24
EEG amplifier. Subsequently, the signals’ noise content was been analysed with analysis
of variance (ANOVA) whereby signal with high noise content was removed from the
samples. Then, spectral energy of different bands of EEG signal (θ, α, β1, β2, β3 and γ)
pertaining to an individual’s visual perception were extracted. Next, dimension reduction
was performed to select band features based on feature separability using Devijver’s
Feature Index (DFI) and Principle Component Analysis (PCA). Finally, neural network
models, namely, multi-layered perceptron (MLP), Elman Recurrent Neural Network
(ERNN) and nonlinear exogenous autoregressive model (NARX) have been designed to
as classifiers to determine the subject’s visual perception, with an average accuracy of
over 90%. Among the trained classifier, ERNN was chosen for it yielded a relatively
higher performance in the both the Locational Matching and Image Recognition
Paradigm in terms of classification accuracies (97.75% and 97.81% respectively).
Therefore ERNN is the most suitable classifier to be used for application of visual
perception to help MND patient navigate in a wheelchair.