Research Of Classification Of Motor Imagery Based On Phase Synchronization And CSP

Brain-computer interface research use will be the focus of informationtechnology in the future. Its research has important implications for rehabilitation andpattern recognition. This paper conducted in-depth study of the following four areason EEG-based multi-category classification of motor imagery BCI system more ideastask control.1)The paper studied processing and extraction of data, elaborated on the basis ofdata extraction, and detailed extraction steps. In addition, the paper set the finalclassification accuracy rate as criteria and thoroughly studied the issue of dataselection period, and ultimately determined the time period from2.5to4.5second forfurther study.2)Based on the idea of multi-class EEG pairwise combinations of EEGclassification, the paper studied the one-to-one and one-to-many CSP algorithms. Thestudy found one-to-one CSP algorithm combined EEG in pairs which then wasobtained by the CSP algorithm. When certain types are found similar, the finalclassification results will be influenced; while on-to-many CSP algorithm took onekind into a class and all the rest another. The resulting pairwise features are going tobe involved in classification. However the former one, while reducing the number ofspatial filters, had little influence on the classification. Considering the aboveshortcomings, the new algorithm presented in this article provides a feasible solution.3)Although an internal spatial pattern selection method of CSP algorithm hasbeen improved, only a spatial filter approach was constructed for direct EEGclassification. This paper brought in-depth analysis into Jacobi approximate jointdiagonalization algorithm CSP. The study found that the algorithm used multiplecategories EEG signal averaging covariance matrix diagonalization airspace to get apreliminary filter, and then use the mutual information theory of optimum spatialpattern selection method to improve. In that case, one can also extract a plurality offeatures for classification. The results showed that the algorithm has reachedmulti-classification purposes, but results are far from perfect.4)Given the shortcomings of algorithms in a multi-category classification, thispaper combined EEG classification and improved spatial mode selection method toclassify two research ideas. Through the analysis, this paper focused on the two theidea of classification and proposed a new algorithm of feature extraction, namely the integration phase synchronization with the CSP inverted binary classificationalgorithms. The essence of the algorithm is that the number of categories of EEG willbe progressively reduced. The algorithm made full use of the features of the index foreach phase synchronization between channels and multi-class EEG optimized pairingcombinations, and then reduced the number of spatial structure in the filter to extractuseful components of EEG categories. It featured a high degree of coincidence weakenthe influence of the two categories of classification results, which were removed insome sense. Only some more efficient features are to be used to improve theclassification accuracy rate that can reach up to82.90%.This paper carried out research for the classification of motor imagery andcombined different research ideas including one-to-one CSP algorithm, one-to-manyCSP algorithm and Jacobi algorithm based CSP approximate joint diagonalizationalgorithm. On this basis, factors affecting the classification results were considered tobring forward a new feature extraction algorithm, namely the integration phasesynchronization with the CSP inverted binary classification algorithm. The resultsshowed that the algorithm could effectively utilize the EEG characteristic componentsto improve the classification accuracy of multi-class EEG.