A Real-time hybrid BCI System for Simultaneous Detection and Classification of the Fusion MI & SSVEP Signals

Abstract

Brain-computer interface (BCI) refers to the use of neural signals acquired by the electroencephalographic (EEG) headset to control the external devices such as the computer, assistive applications and neural prosthetics. BCI manifest gradually diversified in the field of development of an application system. Among them, intention detecting, and remote communication have their unique advantage in numerous areas such as education, self-regulation, production, marketing, security as well as games and entertainment. Besides that, this research in this field has attracted academia and industry alike; the objective is to help severely disabled people to live their life as regular persons as much as possible. However, with a decade of development, this technique still cannot be applied in the daily life, due to its complicated electrode headset, low accuracy, extended training period and non-stationary noise. Based on this background reducing the number of the channel and improving the efficiency is one of the approaches to apply BCI in practical situations easily. Meanwhile, researchers found that every single modality has its defect such as needs train trail, fatigue for the simulation, lack appropriate control conditions, unclear functional significance, etc. A single modality BCI system performs an ID control command which results in less efficient and unnatural movement. Therefore, in this Study, a hybrid EEG (electroencephalograph graphic) system, which simultaneously detects and classifies for the few channels based fusion signal was designed. Motor Imagery (MI) signal and Steady State Visual Evoked Potential (SSVEP) hybrid mental tasks would widely reduce the fatigue for the subjects and extend the control commands (10 commands). Besides, to provide the comfort of wearing the device, our system adapts C3, C4 as the input channels. In here, Blind Source Separation (BBS) approach to get a much clear MI signal and SSVEP signal from the fusion signals. For extracting the feature in each 50ms, Short Time Fourier Transform (STFT) was employed to acquire the features. A 4-layer Convolution Neural Network (CNN) was engaged in this system to distinguish different brain intentions. Besides, Real-time BCI game was designed as the EEG biofeedback using C#, in this game subject can easily control ten commands for the virtual car. Overall, this system proposed a simple structure hybrid BCI system based on two channels, simultaneously detect and recognize ten multi-fusion signals. This kind of the practical and straightforward BCI application would provide the reference towards simplifying BCI field.