Research On Feature Extraction And Classification Algorithm Of Motor Imaging EEG Signal

The emergence of the brain-computer interface(BCI)opens the door to human mind control.At present,the analysis of multiple types of motor imaging electro-encephalogram(EEG)is one of the most important research contents in BCI.However,at present,EEG signal analysis still has problems in feature extraction and low recognition rate.In order to solve these problems,the physiological basis,preprocessing method,feature extraction and classification algorithm of EEG signal are studied.(1)The method of EEG signal acquisition was systematically studied,including the category,organization structure and characteristics of BCI system,the electrode lead method of international standard,etc..Then the basic characteristics of the EEG signal were studied,including the generation,period,frequency and position of the EEG signal.Finally,two more commonly used EEG data sources were selected,specifically the 2005 BCI competition 3A data set and the 2008 BCI competition 2A data set and the experimental examples of these data sets were studied to lay the foundation for subsequent experimental verification.(2)The preprocessing algorithm of EEG signal and the characteristics of four kinds of motion imaging EEG signals were studied and the pretreatment experiment of EEG signal was designed.The experimental content includes the analysis of the pseudo-symbol in the EEG signal,the method of mean interpolation was selected to process the obvious pseudo-recording,and the spectrum analysis of the EEG signal was carried out.And the wavelet packet transform(WPT)method was selected to extract the EEG signal of the specific frequency band which was required.(3)Aiming at the problems of current EEG feature extraction algorithm,such as long consumption and inconspicuous feature differentiation,a three-layer EEG signal feature extraction method based on common spatial pattern(CSP),Hilbert transform and normalization was proposed.It included a first-order spatial feature extraction unit composed of a CSP algorithm,a secondary energy feature extraction unit composed of a Hilbert transform,and an ultimate feature extraction unit finally composed of a normalization algorithm.(4)Aiming at the problems of the low accuracy of EEG classification algorithm,the EEG signal classification model based on convolutional neural network(CNN)algorithm and support vector machine(SVM)algorithm were constructed respectively.And the SVM algorithm was improved by particle swarm optimization(PSO).Finally,the performance of each model was comprehensively tested and analyzed by two sets of offline EEG signal data sets.The algorithm proposed in this paper was compared with other algorithms.The results indicated that the proposed algorithm can effectively extract the features of EEG signals reflecting four kinds of motor imaging,and it was better than other algorithms in recognition rate.Among them,the recognition rate and recognition effect of the SVM algorithm optimized by PSO were the best,and the correct rate was as high as 93.07%.