Research On Multimodal Feature Extraction And Classification Of EEG Signals

Brain-computer interface is a system that can directly command equipment,by analyzing brain signals produced during brain thinking.The development of BCI technology brings new research directions to many fields,and leads to higher demands for researchers.At present,BCI technology has become one of the hot research topics in many fields.Electroencephalogram is an important research signal in BCI system,which can reflect the changes of brain thinking.However,EEG is not a steady signal,nor a linear signal,this is its characteristic,which increases the difficulty of signal analysis.In this paper,the multi-modal feature of EEG signal is taken as the research object,and the two levels of multimodal application in BCI system are deeply explored.Single modal feature extraction and classification are not good,but the issue has been solved to some degree.The key works are as follows:(1)In order to improve the discrimination of EEG signal features,a multi-modal analysis way is proposed,it can combine kinds of domain features of EEG.Firstly,the original multi-channel EEG data is decomposed adaptively by multivariate variational mode decomposition.From the decomposed signal,drawing time domain and nonlinear dynamics features of signals,At the same time,the signal components are combined to construct a new signal matrix.The spatial characteristics of the signal matrix are extracted by using the common spatial pattern algorithm,Then the three features are combined,the multi-modal features of EEG signals are obtained.Finally,for grouping,a support vector machine is used.In comparison with the other four algorithms,the suggested approach is applied in BCI competition II dataset.The experimental results explain,compared with the research of single modal feature,the method has better accuracy,because this method combines the multi-modal characteristics of EEG,including time-domain features,nonlinear dynamic features and spatial features,and achieves better accuracy than single modal features,which proves that the same signal,but modals with have kinds of features is necessary.(2)In order to further improve the function of the system,aiming at the hybrid signal of EEG and near infrared,a multi-modal feature analysis method is proposed.Firstly,for EEG,the wavelet coefficients and CSP features are extracted.In addition,the time domain features and CSP features of fNIRS are studied.Then,the joint sparse representation method is used to realize the fusion and classification of mixed signal multimodal features.The proposed method is tested on a common data set,it is used to collect EEG and fNIRS together.In the experiment,both motor imagery task and mental arithmetic task were explored,it includes the differences and relations between single signal and mixed signal,single modal features and multi-modal features.The findings suggest that better quality is obtained by feature extraction,if in a multi-modal environment.In addition,the use of research methods combined with fNIRS can promote the recognition of EEG signal patterns.In the multi-modal feature extraction and classification of EEG signals,it can provide further reference significance.