Feature Extraction Based On Motor Imagery EEG Signals

Motor imagery(MI)refers to imagining only limb movements without actual limb movements.The EEG generated by motor imagery has the characteristics of event related synchronization(ERS)and event related synchronization(ERD),by analyzing this feature,we can judge the motion intention of the imagination,to realize the control of external equipment.Therefore,motion imagination EEG signal has become the most commonly used feature signal in brain-computer interface systems.In this paper,a real-time online BCI experimental system based on motor imagination EEG signals is completed.(1)A feature extraction method of EEG based on variational modal decomposition and AR model power spectrum estimation is proposed.After the original EEG signal is decomposed by variational mode,it is sent to the AR model for calculation,the power spectrum of the calculation results is estimated,and finally,the eigenvalues are sent to the support vector machine for classification.Compared with the traditional single feature extraction method,it has more advantages.(2)An EEG feature extraction method based on optimal variational modal decomposition and approximate entropy is proposed.Aiming at the problem that the K value of variational modal decomposition can not be taken automatically,this paper uses the Sooty Tern Optimization Algorithm and automatically search for the best K value.The decomposed components are calculated by approximate entropy,and the appropriate features are selected from the calculation results for classification.This method significantly improves the recognition efficiency.(3)A feature extraction method of EEG based on variational modal decomposition and depth belief is proposed.The classification results of most existing models depend on the richness of EEG spatial information to a certain extent.In fact,data acquisition with fewer channels is more convenient.Therefore,how to extract and classify EEG signals is an important problem in the case of insufficient spatial information.Therefore,this paper proposes to use variational mode decomposition to decompose EEG signals,and then use Hilbert transform to cover the time-frequency domain characteristics of complete Mi time history,Marginal spectrum(MS),instantaneous energy spectrum(IES)and joint feature timefrequency(JT-F)are extracted and integrated.DBN is introduced to reduce the dimension of the fused high-dimensional features to obtain the optimal features of motor EEG patterns,realize the recognition of motor imaginary EEG patterns,and avoid the problem of reducing the recognition rate caused by manually determining the optimal time period and optimal frequency band.