Motor Imagery EEG Processing Based On Common Spatial Pattern

Brain computer interfaces(BCIs)technology applys computer equipments to read electroencephalogram(EEG)when the human's brain is thinking,then via the reading of EEG to communication with the external environment,evading the original human peripheral nerve and muscle system.BCIs,a new type of human-computer interaction,shows its practical value in many fields.Therefore,it has become a hot topic in many interdiscipline recently years.At present,three types of EEGs,motor imagery EEG,P300 and steady-state visual evoked EEG,are commonly used in BCIs system.P300 and steady state visual evoked electroencephalograms are produced by the stimulus patterns of external environment,and motor imagery is entirely spontaneous,so the application prospect of motor imagery is more widely and the implementation difficulty are also relatively high.Recognition of EEG signals is the core part of brain computer interface technology,therefore this paper focuses on motor imagery EEG,showing a detailed study of the EEG signal processing methods,including preprocessing methods,feature extraction algorithm,classification algorithm,and design system of the offline and online BCIs.In view of the above aspects,this paper has done the following work:(1)Realization of EEG signal acquisition,feature extraction and simple analysis;(2)The methods of extracting and processing the EEG signals of motor imagery are studied,including spatial filtering,frequency band energy,Fourier transform,wavelet packet decomposition,common spatial pattern and adaptive parameter regression method;(3)I also study the linear classifier and support vector machine classifier,and verified the effectiveness of the two classifiers through experiments.A classification method suitable for on-line system is found;(4)In this paper,wavelet decomposition and common spatial pattern are combined to extract the feature of EEG signals,and the validity of this method is verified by experiments.An efficient feature extraction method is proposed;(5)According to the EEG signal energy distribution in the cerebral cortex for four class motor imagery,I proposed the EEG signal recognition method of multi-class motor imagery based on feature reconstruction and wavelet transform,and this method is tested by competition data set;(6)I designed a new type experimental paradigm,and compared it with classical paradigm's result,found that different visual display have different effects on subjects EEG signals,so the research on improvement of experimental paradigm is also very necessary;(7)Applying the improved CSP algorithm,supervised learning method,the new experimental paradigm and other technical factors,I brought about a on-line brain computer interface system for recognition of EEGs by programming,and achieved higher recognition accuracy.