EEG Signal Processing And Classification For "Imitating Reading" BCI

Brain-Computer Interface (BCI) technology is not dependent on brain’s normaloutput channels such as peripheral nerve and muscle tissue. It provides a directcommunicate method between brains with the outside world for the patients whosenerve or muscle tissue has damaged that can not use normal means of communication,in order to achieve the control of external devices such as computers, to improve thequality of life of patients. In addition, the Brain-Computer Interface technology in thefield of human-computer automatic control and military also has potential researchvalue.The core issue in the Brain-Computer Interface system is a translation algorithm.That is how to convert electrophysiological signals from the user (such as EEG) tothe output control signals to control external equipment.In this thesis, through thereviewing the literatures and summarizing the status quo of the internationalBrain-Computer Interface research,as well as a variety of EEG signal processingmethods,we study the points from the following.(1) Experimental model of EEG acquisition and pretreatmentThe experimental model used the acquisition mode based on the“Imitatingreading”Evoked Potentials by the Biomedical Engineering Laboratory of CentralSouth University for Nationalities. The subjects of the mode of physical stimulationwithout the mutation target symbols as usual read access to visual information.Because the EEG signal is very weak and extremely vulnerable to the impact of hotnoise of various kinds of interference,so in this paper we can reduce eye movementartifacts interference pollution by using the experimental model based on“Imitatingreading”, and then use low-pass filtering, down-sampling to further filter outinterference and noise pollution of EEG. The experiments show that the results aresatisfactory.(2) The extraction of EEG characteristics.Effectively extract the brain thinking activities in the consciousness ofinformation is one of the key technologies of brain-computer interface research andis the basis to correctly identify of the different modes of consciousness. In referenceto many EEG feature extraction method based on a common spatial patterns (theCommon Spatial, Pattern,CSP) method after pretreatment EEG feature extraction, themethod the goal is to design spatial filter are two categories in the high-dimensionalspace, to find one to maximize the first-class variance while minimizing the secondclass variance. The experimental results show that it has significant results for the subsequent classification and recognition by using this method to extract the EEGcharacteristics.(3) The design of EEG classifier.The classifier design is another important part of the Brain-Computer Interfacesystem, the performance of the classifier has a direct impact on the performance of theBrain-Computer Interface system. In this paper, using support vector machine (theSupport Vector Machines,SVM) method for classification of EEG, the algorithm isbased on statistical learning theory, taking experience minimization principle as aprecondition, seeking a optimal surface which not only can accurately separate thetwo types of sample data, but also can maximize the classification interval betweenclasses. The experimental classification results show that the method of classificationand identification of the EEG has a good effect.Finally the EEG based on“Imitating reading”was classified by the processing ofthe above steps, the results showed that the classification and identification of theresults are satisfactory. In this paper, we performed experiments of five subjects andthen classified the EEG got form them. Each subject was performed randomized15repeated trials experiments to get every classification accuracy rate and the averageaccuracy rate. In this paper, we performed experiments of five subjects and thenclassified the EEG got form them. Each subject was performed randomized15repeated trials experiments to get every classification accuracy rate and the averageaccuracy rate. The results showed that the highest classification rate can reach97%and the average accuracy rate can reach more than90%. Therefore it can be drawnthrough the experimental results that the method combined of CSP and SVM issuitable for the study of EEG based on “Imitating reading”.