Research On Artifact Rejection Methods Of EEG Signal For Brain Computer Interface

Brain-computer interface(BCI)provides a way to develop interaction between a brain and a computer.The BCI technology based on Electroencephalograph(EEG)is most commonly used.In recent years,there has been a rapid development in the fields of medical,rehabilitation,military,aerospace and entertainment.The portable BCI system has become the key to bring BCI technology to practical applications.But the devices of portable BCI rely on artifact rejection methods,because they are easily effected by artifact interference and the movement of BCI users.Independent component analysis is one of the most commonly used methods which can separate artifacts from EEG signals effectively.However,it needs professionals to handle the selection of components so that it is not a convenient method to implement to the practical application.Therefore,this study proposes an automatic artifact rejection method based on pattern classification,which is on the basis of ICA.Our study mainly includes:(1)By studying the experimental settings and signal characteristics of different BCI systems based on P300 evoked potential and SSVEP,the EEG components and artifact components are obtained after comparing and rejecting the independent components manually.In order to obtain better artifact components,we used three different ICA methods to separate artifact components from EEG signals,including Infomax,Jade and FastICA,and then rejected the artifacts manually according to different physiological characteristics of components.The result shows that we obtained a set of artifact components by all methods of ICA as well as the effect of artifact rejection for EEG signals is significant.(2)This study extracted 10 different features in temporal,spectrum,and spatial domain from components of P300 EEG signals as the training samples of automatic artifact classifier based on Bayesian Linear Discriminant Analysis.After the artifact rejections of P300 and SSVEP,the artifact components were separated from the EEG signals in general and the collection of artifact components were similar with manual artifact rejection by comparing the results.It is proved that the automatic artifact classifier is adaptive to different types of EEG signal,and the 10 features extracted from different domains are the common properties of artifact components.(3)We also used the automatic artifact rejection method on P300 data of BCI competition III and SSVEP data of high speed SSVEP-based BCI before the classifications.The accuracies and information transmission rate of classification after artifact rejection are obviously improved compared with those before artifact rejection,which verifies theeffectiveness of the automatic artifact rejection method in improving the classification performance of BCI systems.According to the above research content,the innovation points of this study includes,by studying different BCI systems,it compares the performance on components separation of several ICA methods.Last but not least,this study provides an idea for automatic artifact rejection method designing based on feature extraction and pattern classification,and verifies the importance of artifact rejection method on improving the performance of BCI system.At the same time,it draws the conclusion that ICA,as an artifact extraction algorithm,can not separate artifact components from EEG completely and points the research direction of artifact rejection method in the future.