Optimization Of Artifacts Removal And Intention Recognition Bsaed On Motor Imagery Eeg Signals

Motor Imagery(MI)is one of the most commonly used control patterns in Brain Computer Interface(BCI)systems.The classification and recognition of electroencephalogram(EEG)signals is an important aspect of MI-BCI system.In this paper,the left or right hand motor imagery of EEG signals are taken as the research object,to improve the classification accuracy of the EEG signals.Combined with the characteristics of the EEG signals,the process of artifact removal,feature extraction,classification and recognition of the EEG signals are analysed and studied.The main work is as follows:(1)Firstly,three kinds of motor imagination EEG data sets are described in detail:simulation EEG data,BCI 2008 competition EEG data,and self-collected EEG data set in laboratory.The construction of simulated EEG data used to evaluate the performance of the artifact removal method is emphasized.The BCI 2008 competition data set and the EEG data set collected by the laboratory can verify the feasibility and effectiveness of the EEG classification model.(2)For different kinds of artifacts,such as power frequency interference,baseline drift and biological artifacts,different methods are adopted to remove them.Among them,a hybrid algorithm based on multi-empirical mode decomposition and independent component analysis was selected to remove the electroencephalography artifact which is the most difficult to remove.In view of the fact that different artifact removal orders will affect the final classification results,an optimization strategy for artifact removal is proposed.The signals traversal all the combined paths,and the simulated EEG data were evaluated by relative root mean squared error and correlation coefficient.The measured data were further verified by power spectral density,and the optimal removal order was obtained by comprehensive evaluation.It lays a good foundation for the following feature extraction and classification.(3)To solve the problem that it is difficult to extract the optimal time-frequency features of EEG signals due to individual differences,a feature extraction and classification method based on convolutional neural network is proposed.Using CNN network automatic learning deep feature for feature extraction and classification,considering the small sample size of EEG signals,the EEG signals after preprocessing get feature of energy,PSD and fusion by feature extraction,these features subsets are respectively input into the convolutional neural network for feature extraction and classification,and finally the boosted CNN model is established through weighted voting.Compared with the traditional EEG signal classification and recognition algorithm,the classification accuracy of this method is improved.