Research Of Paradigm And Classification Algorithm Of Brain-computer Interface Based On Face Distinguishing

In the past few decades,the brain-computer interface system has been developed to achieve stable performance.However,due to the poor practicality of this system,it still can’t be invested in large-scale applications.On the basis of summarizing,this paper improved the performance of P300-brain computer interface system from two aspects.First,to optimize the paradigm,which is,to propose a paradigm that can induce event-related potentials with greater amplitude and more types.Second,to optimize classification algorithm,which is,to explore a classification algorithm with better training model and high classification accuracy.For the paradigm,to explore the cognitive mechanism of face distinguishing and the influence mechanism of familiarity on BCI system,based on the famous face paradigm,we proposed to replace the famous face(Chinese basketball player Yao Ming)with the familiar face(mother)and the own face,and used the matrix transformation to avoid the “neighboring intensify” phenomenon.Exploring the characteristics and differences of the event-related potentials induced by three face paradigms,the performance of three paradigms were compared by Bayesian linear discriminant analysis classification algorithm.The results showed that the own face paradigm achieved greater amplitudes of P300 and higher classification accuracies than the familiar face paradigm and famous face paradigm.As for the classification algorithm,we studied the application of deep learning for EEG classification,and proposed to apply convolutional neural network to classify EEG.A convolutional neural network model for EEG classification was designed and implemented.The EEG data that collected in the three paradigms experiments were trained and tested by this CNN model,then compared the results with the traditional BLDA algorithm.Both results showed that the classification accuracy of own face paradigm was higher than the other two face paradigm.However,due to the type and amount of data and the parameter setting,the CNN didn’t achieve better classification accuracy than BLDA.In summary,compared to the familiar face paradigm and the famous face paradigm,the own face paradigm improves the performance of the P300-BCI system.Besides,the CNN applied for EEG classification needs to be further optimized.