Research On Eeg Intention Recognition Based On Sparse Common Space Pattern And Regularized Discriminant Analysis Method

Brain-Computer Interfaces(BCI)are systems that enable their users to skip the peripheral nervous system and muscles,and only need to collect signals from the brain to communicate with computers or other devices.And the active electroencephalogram(EEG),such as motor imagery EEG,classification is an important issue in the BCI system.However,it is still necessary to collect multi-channel EEG and the recognition accuracy is often difficult to meet the requirements.Preprocessing,feature extraction and classification can affect the entire classification result in EEG processing.In this paper,motor imagery EEG are taken as the research object.The best preprocessing method is determined by experimental analysis,and the sparse feature extraction and regularization discriminant analysis method are proposed to improve the classification performance of the discriminant system.The main contents of the paper are as follows:(1)The public data set and the independent experiment data set are obtained by consulting the literature and independently designing the experiment,and the study was carried out under the two data sets.Signal preprocessing often has an important influence on the classification of the signal.Therefore,the paper validates different preprocessing methods by the two data sets,and the best preprocessing method were select based on the results to filter the original EEG signals accordingly.(2)A novel feature extraction algorithm,sparse common spatial pattern(SCSP),is proposed to solve the problem of lack of effective strategy in data selection of each channel and spatial filter in multi-channel EEG classification and recognition.The SCSP algorithm can effectively overcome the problem that the feature vector space extracted by the traditional common spatial pattern(CSP)algorithm will repeatedly select the feature mode,and the difference of the extracted characteristics is more obvious.The experimental results of the public data set BCI competition IV data set I also verified the effectiveness of the improved algorithm.(3)To solve the problem of singular values encountered in matrix decomposition that cannot be handled by traditional linear discriminant analysis methods,the regularization discriminant analysis is proposed by adding regularization parameters,which overcomes the deficiencies of linear discriminant analysis and improves the accuracy of feature classification.The experimental results on the public data set BCI competition IV data set I and the autonomous experimental data set all have proved the effectiveness of the improved algorithm.