Electroencephalogram Classification Based On Motor Imagery

Scientists believe that in the new century,we will make breakthroughs in brain science and cognitive neuroscience.On the study of brain-computer interface signal transmission rate is slow and eeg signals low accuracy of two kinds of problems,this paper respectively in two classification of eeg and four classifications on brain electrical solution,two categories and four categories respectively in the 3rd contest of eeg data sets a and the third eeg series data set in the three a verification,and than the eeg competition first prize in the accuracy of the data set.For the first time,S transformation(Stockwell transformation),genetic algorithm and BP(Back propagation)neural network were adopted in the dichotomy,and good results were obtained.In feature extraction,S transformation is adopted,and the gaussian window of S transformation is locally scalable and mobile,which is a feature that the continuous wavelet transform does not have.The S transform establishes a relationship with the Fourier spectrum,and the resolution it provides is frequency dependent.Characteristics of optimal selection genetic algorithm(ga),chose the Back propagation neural network classification algorithm,using genetic algorithm to optimize the BP(Back propagation)neural network,BP neural network need not set random weights and thresholds,in BCI competition III data sets are verified in the I,with 94.5% accuracy,higher than that of eeg series the first 91% of the 3.5% accuracy.Second chapter,based on binary classification research of eeg signals is divided into two parts,the first part is based on S transform and electrical characteristics,the second part USES genetic algorithm to find the weights and threshold of BP neural network,and in ten layer of neural network are classified,and in the data set on the verification of accuracy more than most of the algorithms.Four classification in the feature extraction using the PW-CSP(pair-wise CSP)and the Hilbert transform,the choice on the classification of SVM(Support vector machine),the experimental results in the 3 a BCI competition III is verified,in k3 b participants accuracyreached 94.4%,the l1 b subjects on accuracy reached 83.3%,the average accuracy rate reached88.85%,electrical contest the highest accuracy of 78%,improved the accuracy of10.85%.Chapter iv based on the four classification research of eeg signals is divided into two parts,the first part is the arithmetic sum characteristic signal,the second part is a single fine classification process,the support vector machine(SVM)method is simple,time overhead is small,practical,simple program is running,and in the data set on the verification of accuracy more than most of the algorithms.