Limited Penetrable Visibility Graph And Deep Learning For SSMVEP EEG Signals Classification And Fatigue Mechanism Characterization

Brain Computer Interface(BCI)based on Steady-State Motion Visual Evoked Potential(SSMVEP)can relieve discomfort caused by strong visual stimuli of SSVEP.SSMVEP merges the motion perception capabilities of the human visual system,which has attracted a great deal of attention.As an unavoidable problem,the mental fatigue state that occurs after a long time use of the SSMVEP-based system directly affects the efficiency of the BCI system.In order to characterize the fatigue behavior of the brain and improve the general applicability and the classification accuracy of SSMVEPbased BCI system,a series of studies have been carried out in this dissertation.In order to alleviate the strong visual stimuli caused by SSVEP,we design a SSMVEP stimulation paradigm interface based on Newton’s ring.Offline experiment and online experiment are designed,and the effectiveness of the interface is proved by double validation.The spectral energy map of the EEG data collected by offline experiments visually shows that the designed Newtonian ring paradigm can induce SSMVEP signals with significant frequency domain characteristics.The online experiment results show that the designed Newtonian ring paradigm can achieve precise control of external equipment.In this dissertation,we propose a multiplex limited penetrable horizontal visibility graph(MLPHVG)-based method to characterize fatigued behavior based on SSMVEP signals.We design the SSMVEP-based fatigue experiment and obtain the signals of the subjects under normal and fatigue conditions.We classify the signals via the CCA-SVM method and the results show that the accuracy rate of subjects under the fatigue state decrease significantly.Brain networks in two mental states are constructed using a complex network approach based on MLPHVG.Then we calculate the average weighted clustering coefficient and the weighted global efficiency of the brain networks generated under two different states to characterize the fatigue mechanism of brain under SSMVEP paradigm.The results suggest that when the brain state changes from normal to fatigue,the average weighted clustering coefficient and the weighted global efficiency corresponding to these two brain states decrease,and the associations and information transfer efficiency among different brain regions become weaker.Our analysis sheds new insights into the understanding and management of the fatigued behavior using the SSMVEP-based BCI system.In order to improve the classification accuracy of SSMVEP signal under fatigue state,we designed two kinds of convolutional neural network,i.e.the classic convolution neural network and the expansion convolution neural network based on the grouping module.Based on the pre-processed SSMVEP signals,the features in frequency domain are extracted by FFT transform,and two kinds of convolutional neural network are used for feature fusion and classification.Then,we compare the proposed methods with the CCA method.The results show that the proposed method can improve the classification accuracy under fatigue state to a certain extent,and the expansion convolution neural network performs better.