Research And Design Of Brain-computer Interface System Based On Event-related Potentials

Brain-computer interface (BCI) is a new information exchange system, which candirectly make communication and control between the human brain and the computers,without requiring people’s languages or actions. Brain computer interface technologyis related to the intersection of several disciplines, for example, neurological,psychological cognitive science, rehabilitation engineering, biomedical engineeringand computer science and so on. Many methods are used to research the BCI system,such as the electroencephalogram (EEG), cortical electroencephalogram (EcoG),functional magnetic resonance imaging (fMRI), functional near-infrared opticalimaging (fNIR).In this paper a BCI system is researched and designed based on Event-RelatedPotentials (ERP), after researching and analyzing many papers about BCI system athome and abroad in recent years. A new input system is designed based onvent-related potentials, using non-invasive EEG-based BCI research methods. It canbe used as a password input system without hands and directly controled by the EEG.The contents of the researches include the whole input system, such as experimentalparadigm, the design and realization of the system, the analysis and processing of theEEG data.The experimental paradigm is designed by comparing the form and color of theinput characters according to the P300potential characteristics. Based on this, manyexperiments are performed with different parameters. The input system is realizedmainly by programming on the VC++MFC platform. The system combinesgeneration of the stimulating interface, collection, restoration and analysis of the EEGdata, making it available to output the target character watched by the subject on thedisplaying interface in real time.The processing scheme of the collected EEG data is determined by analyzing theP300data in the BCI III. First, for the competition data, a digital filter is used for filtering and down-sampling and time windows are used for dimensionality reduction.Then the classification of the EEG data is followed by five methods, including Fisherlinear discriminant analysis (FLDA), Regularization Fisher, Kernel Fisher, SupportVector Machine(SVM) and Bayesian linear discriminant analysis (BLDA). The resultsshow that Regularization Fisher superimposed within a shorter time using fewernumbers of times to reach the highest correct rate, the Fisher second. So theself-collected EEG data can be classified by Fisher and Regularization Fisher afterremoving the noise, which can be analyzed and processed by average, IndependentComponent Analysis (ICA). So that the system has achieved the best way of input,experimental parameters and the scheme of EEG data processing.