Indian Journal of Science and Technology

Abstract

Electroencephalography (EEG) signal is a biological signal which can be associated to the mental task of a person. A Brain-Computer Interface (BCI) can be designed such that the mental activity accounted for the visual perception of a person can be recorded and subsequently converted into a control signal for controlling the movement of a wheelchair. Comparison was made between the Multi-Layered Perceptron feed-forward network (MLP) and Nonlinear Autoregressive Exogenous model (NARX) as a variant of Recurrent Neural Network (RNN). The networks were designed to discriminate the different brain activities when the subject was being presented with different visual stimuli. The trained network models have yielded an average accuracy of 93.3% for MLP models and 98.1% for NARX models.
Keywords: Brain Computer Interface (BCI), Electroencephalography (EEG), Levenberg-Marquardt Training Algorithm (LM), Multi-Layered Perceptron Neural Network (MLP), Nonlinear Autoregressive Exogenous Model (NARX), Power Spectral Density (PSD), Visual Perception