Research On EEG Motor Imagery Based On Deep Temporal-domain Information Extraction And Its Application

Brain Computer Interface(BCI)is a technology that recognizes people’s intentions through neural activity and converts electrophysiological signals into device control commands.Electroencephalography(EEG)based motor imagery(MI)Brain-Computer Interface has been widely applied in constructing a pathway between human brain and external devices.However,since EEG severely suffers from low signal-to-noise ratio and unpredictable pattern variation,the decoding of MI-based EEG signals remains a challenging task.Therefore,this thesis proposes an EEG motor imagery classification model based on deep temporal-domain information extraction and a graph embedding representation algorithm,and designs a hand exoskeleton control system.The specific work are as follows:1.This thesis proposes an EEG motor imagery classification model based on deep temporal-domain information extraction: DTC-EEGNet.To fully extract the spatial and temporal information of EEG for MI decoding,this thesis designs an end-to-end deep convolutional neural network model,combining EEGNet and a temporal convolutional network.The proposed model requires few data pre-processing and a small number of trainable parameters.Compared with other four EEG-MI recognition methods on three datasets,the proposed method significantly improves the classification performance.2.This thesis proposes a graph embedding representation algorithm.Based on the spatial topology of EEG electrodes,two undirected graphs and adjacency matrices are defined,and then the normalized adjacency matrices are embedded into the EEG signals according to the spectral graph convolution theory.The experimental results show that the classification accuracy of both graph embedding representation methods is higher than that of all contrasting methods.The graph embedding method can enhance the brain region representation ability of EEG signals and solve the problem of missing values in EEG channels.3.This thesis designs an experimental paradigm of motor imagery to construct a laboratory dataset and a hand exoskeleton control system.Experimental results show that proposed method achieves the highest classification performance on the laboratory dataset.The online experiment of the hand exoskeleton control system provides a new idea for the development of the MI-BCI system.