Research On Adaptive Brain-Computer Interface Control System

Brain-Computer Interfaces (BCIs) provide a new communication and control channel that does not depend on brain's traditional output pathways of peripheral nerves and muscles. An applicable BCI control system must have the ability of self-regulation to adapt to the time-varying electroencephalogram (EEG) signals. Consequently, to maintain robust performance, the signal processing methods should be able to monitor the change of EEG signals and adjust the parameters online. Based on the above consideration, this paper investigated the algorithms of adaptive feature extracting and classification, and applied them to the BCI control systems designed for virtual inverted pendulum and automatic-car control.The EEG patterns of left and right hand motor imagery were used in the virtual inverted pendulum control experiment. Online training paradigm was performed to make the subjects execute motor imagery tasks more practicedly and enhance the quality of the signals. The BCI control system in this experiment ultilized Adaptive Common Spatial Pattern (aCSP) and adaptive Linear Discriminant Analysis (aLDA) algorithms and the results showed that, the proposed control system could achieve self-regulation according to the change of EEG signals, increase the classifier's accuracy, and prolong the control time of inverted pendulum.The other experiment was automatic-car control experiment which applied left, right hand motor imagery and two hands relaxation as input EEG patterns. On the basis of aCSP and aLDA, new feature extracting and classification algorithms were designed for three-class issue. Simultaneously, a sliding-window method was utilized to enhance the system's accuracy and stability. The experiment results showed that, the BCI comtrol system based on these algorithms obtained satisfying performance.