Executive Function Training System And EEG Signal Analysis Based On BCI-VR

Executive function is a high-level cognitive activity of human brain.With the gradual improvement of the knowledge system of brain-computer interface and cognitive rehabilitation training,the cognitive training evaluation methods for executive function are not limited to the form of questionnaire scale.However,the cognitive evaluation process is boring and lacks of deep participation,which leads to the inaccurate evaluation of executive function.Therefore,based on the theoretical knowledge of executive function and braincomputer interface,this thesis designed an executive function training system,which combined the behavioral data and EEG signals before and after training for systematic evaluation.First,under the guidance of the theoretical knowledge of executive function,the braincomputer interface and virtual reality technology are integrated to construct the executive function training system based on brain-computer interface and virtual kitchen.Fifteen subjects were recruited for 35 days of executive function training.During this period,EEG and behavioral data of the subjects were collected,and T-test statistical analysis was subsequently used to evaluate the time consuming and score of the tasks in the behavioral data.Feature extraction and classification of EEG signals were carried out using rank condition mutual information and random forest.Secondly,a multi-scale separable residual network model is proposed to classify the EEG signals with respect to executive function.In this network,multi-scale convolutional neural network is introduced to accurately analyze and capture the features of executive functional EEG signals from multiple scales,and then features are further obtained from time domain and space domain by depth-separable convolution.In addition,residual connection is used to improve the robustness and convergence speed of the model.Finally,the changes of executive function before and after training were evaluated from two aspects of behavioral data and EEG signal,and the effectiveness of the training system was verified.The superiority of the multi-scale separable residual network was verified by comparing other EEG classification algorithms and combining various classification indexes.The effectiveness of the system was further verified by the analysis of the results.