Recognition Of Brain Computer Interface Instruction Based On P300

A Brain-Computer Interface(BCI)is a communication system that analyzes brain signals of thought and translates them into commands for external devices or program.The purpose of the BCI is to provide a new way of communication for those who are damaged or motor disabilities.In the field of BCI research,BCI based on P300 has the characteristics of simple experimental paradigm,subjects do not need to be trained and can generate many instructions,which is favored by many researchers.Therefore,this paper mainly studies the instruction recognition based on P300's BCI.The main contents of this paper are as follows:(1)Two kinds of channel selection index algorithms are proposed.One is the Channel Sensitivity(CS)based on the amplitude of the EEG signal.The index takes the absolute value of the difference between the P300 and the non-P300 waveform of each channel as the criterion of channel selection.The other is the Phase Synchronization(PS)based on the phase characteristic of EEG,which takes the size of PLV value between channels and channels as the criterion of channel selection.The experimental results show that the algorithm can choose different optimal channel configurations according to different subjects and reduce the amount of computation.(2)Two different time domain feature extraction methods are studied and implemented,which are superposition average method and the principal component analysis method.Finally,the feature selection of the principal component analysis method is explained in detail and the feature extraction results of the P300 EEG signal superimposed 15 times are shown.(3)The Bayesian linear discriminant analysis(BLDA)method based on probability model is studied.This method is used to verify the effectiveness of feature extraction and channel selection algorithm.The results of BCI Competition Data Set II show that the use of all channel data,a feature extraction method of principal component analysis and classification for the case of BLDA,the character recognition accuracy rate reaches 96.5% and the first prize of the contest results are equal.The accuracy of 8 optimal channels based on PS index is 67%,and the accuracy rate based on CS index has reached 84.5%,which is higher than that of 8 fixed channels.(4)A classification algorithm of weight extreme learning machine is proposed.In this method,the random function of the ELM algorithm is not associated with the sample class and the P300 positive and negative sample is not balanced.Finally,on the same experimental set,we verify that using all channel data,feature extraction method and WELM classifier,the accuracy of character recognition reaches 97%,while the 8 channel data using CS index achieves 85.5%.