Recognition Algorithm Of Four-class Motor Imagery EEG

The brain-computer interface(BCI)system is a communication system for controlling a device,by human intensions,which does not depend on the brain’s normal output pathways but relies on the detectable signals representing responsive or intentional brain activities.Electroencephalograph(EEG)signal contains rich effective information.So,they acquire and analyze of the EEG signal,then extract and recognize evoked EEG features of motor imagine,and human brain can transfer the information to the outside world and control the external devices effectively.This provides a new way to communicate with the outside world for people who have normal brain’s thinking but lost motor function.BCI’s research has important potential applications in many fields such as artificial intelligence,rehabilitation engineering,military and aerospace.The pattern recognition of EEG motor imagery signal in BCI has been studied at domestic and overseas.There has a higher accuracy in EEG pattern classification of the left and right hand movement imagination at present.However,the problems of low accuracy of classification,algorithm complexity and poor stability are still existed in classification of multi-task motor imagery of EEG signals,and its wide applications are limited.For the problems of four-task motor imagery of EEG signals,the paper is focused on the methods of the acquisition,preprocessing,feature extraction and classification of the EEG signal evoked by imagery motor.The main research work is shown as follows:(1)Acquisition and processing of EEG data and selection of electrode leads.Using the motor imagery experiment paradigm on BCI international competition,by the 64 leads electroencephalograph of Neuroscan,they collected the EEG datasetⅢa of 2005 BCI and selected lead.Then,the coherent average method is used to preprocess the EEG signals.This method can remove the spontaneous EEG signal,and retain EEG signal evoked by multi classification motor imagery.(2)Extract features of EEG signal evoked by imagery motor.The paper proposed three methods to wavelet transform,improved common spatial pattern(CSP)and wavelet-common spatial pattern analysis for the feature extraction of EEG evoked four kinds of imaginary motion.Then,using the"one-versus-rest" method of CSP,they constructed multiple CSP filters,get projection signal of each kind of motor imagery modes by CSP mapping.Then,obtain feature vectors by calculating variance to the projection signal energy and doing logarithm operation.After the difference processing to the vector,EEG features for different motor imagery modes were extracted.Then,EEG features are used to optimize the electrode leads.The simulation results have shown that wavelet-common spatial pattern method can effectively perform feature extraction for four classification imagery motor.(3)Fusion decision classification of EEG features for motor imagery.The support vector machine(SVM)based on radial basis kernel function,optimization parameters of grid optimization method,is used to carry out four-task classification,combining with the CSP to classify the feature signal of EEG.Computer simulation results show that the average accuracy rate of classifying is 90.9%for maximum,the Kappa is 0.867,and the algorithm execution time is shorter.The method is a better classification to a four-class motor imagery of EEG signals.(4)Research of nonlinear classification of EEG features for motor imagery.Due to the strong nonlinear and non-stationary characteristics of EEG signals,and the probabilistic neural network is suitable for processing four-class motor imagery of EEG signals.Computer simulation results show that the correct rate of classifying is 92.31%for maximum and this method can be applied to multi-class motor imagery of EEG signals.The results of the study can be applied to the BCI technology and intelligent prosthetic system,also can provide the basis for the study of brain cognitive activities and motion control,It has the double value of science and application.