The Research Of Multivariate EEG Signal Analysis And Its Application In BCI

Brain-computer interface technology directly starts from the source of the human mind–the brain,and establishes a bridge between the brain and computers or other devices.The technology subverts the traditional way to transport information which depends on humans' peripheral nerves and muscles.Since the technology has broad application prospects in many fields,such as medicine,entertainment,intelligent life and so on,it has become one of research hotspots of brain science.In this paper,the aspects of preprocessing,feature extraction and pattern classification on multivariate motor imagery EEG signals are studied,and applied to control electric prosthetic based on EEG signals.The main research work and achieved results are given as follows:(1)Preprocessing: In order to reduce the interference of noise,this paper adopts two kinds of preprocessing methods: Butterworth filter and adaptive wavelet threshold de-noising method.Experiments are carried out on the public competition data set 1 in 2008.The result shows that the former can obtain the frequency band information related to the motor imagery rhythm signals,and the latter can effectively reduce the noise interference to improve the signal-to-noise ratio.(2)Feature extraction: Aiming at the problem that selecting IMFs depends on experience,a novel identification method of information-bearing IMFs is proposed based on noise-assisted multivariate empirical mode decomposition(NA-MEMD)and mutual information,and applied for feature extraction of EEG signals.Firstly,multi-channel EEG signals are decomposed by the NA-MEMD algorithm to obtain the IMFs at each scale.Secondly,mutual information is used to calculate the correlation between cross-channel EEG signals and their IMFs,noise signals and their IMFs,EEG signals' and noise signals' IMFs,respectively,obtaining the sensitive factor of each scale used to recognize the useful IMFs which are then used to obtain the reconstructed signals corresponding to each channel by adding them together.Finally,the common spatial pattern(CSP)approach is employed to extract features of the reconstructed signals.The useful information related to EEG signal is selected self-adaptively,which can effectively improve the degree of feature extraction.The effectiveness of the proposed method is verified by comparing with other selection methods.(3)Pattern classification: The conjugate gradient method is used to determine the parameters in the traditional Gaussian process.However,the conjugate gradient method has a strong dependence on the initial value and is easy to fall into local optimum in optimization effect.In order to solve the problem,this paper proposes a Gaussian process classification(GPC)method based on artificial bee colony(ABC)optimization,which is used for pattern recognition of EEG signals.Firstly,Gaussian process model is constructed,and suitable kernel function is chosen and the parameters to be optimized are specified.Then the reciprocal of the recognition error rate is selected as fitness function,and the parameters which are used to obtain optimal accuracy in a limited range are found out by employing the ABC algorithm.Finally,the Gaussian process classifier with optimized parameters is used to classify the samples,and the effectiveness of the proposed algorithm is demonstrated by experiments.(4)The research of multivariate motor imagery EEG in the control of electric prosthesis: Firstly,the experimental background is introduced.Secondly,the overall control scheme is designed,and the acquisition experiment paradigm is designed and performed.Thirdly,the adaptive wavelet threshold de-noising method is used for preprocessing.The NA-MEMD and mutual information are used to extract the EEG signals' features.The ABC-GPC method is used to classify the extracted features.Finally,the classification results are mapped into the corresponding control commands,driving electric prosthetic to execute the motions of hand extension and hand grasp.