Research On Cross-subject Emotion Recognition Based On Novel Flexible Fabric EEG Monitoring System

EEG is closely related to people's emotional and cognitive activities,and can reflect the changes of emotion.Compared with other physiological signals,EEG has better time resolution,so it is widely used in emotion recognition.In order to better monitor EEG and perform emotion recognition,people have put forward higher requirements on EEG acquisition equipment,which not only requires high reliability of the collected signals,but also requires the equipment to be comfortable to wear and easy to carry.The researches of emotion recognition based on EEG mainly have the following problems:(1)Traditional EEG acquisition equipment often needs to wear an EEG cap,which is cumbersome to use and not suitable for long-term EEG monitoring;(2)There are few studies on emotion recognition based on three-lead EEG from the forehead,and the EEG features are not fully explored;(3)EEG based emotion recognition has the problem of large individual differences across subject,and the accuracy of the classification is not high.Based on the above problems,this dissertation mainly carried out the following three tasks:(1)The dissertation designed a flexible fabric EEG monitoring system,which is composed of flexible EEG dry electrodes,EEG data acquisition circuit and display software,and carried out open and closed eyes and SSVEP test experiments.EEG collected by the system accurately reflects the alpha rhythm of the brain.Under 12Hz visual stimulation,the signal produces a response related to the stimulation frequency,which verifies the reliability of the signal collected by the system.It is comfortable and convenient to wear and suitable for long time EEG monitoring.(2)The dissertation established a sparse lead EEG dataset that can be used for emotion classification,and conducted research on positive and negative emotion classification.Based on the EEG monitoring system of the novel flexible fabric,the dissertation collected the EEG of 14 subjects under the evoked virtual reality emotional scene,and extracted the EEG features such as wavelet entropy and wavelet energy,and then input the feature matrix into XGBoost.And the accuracy reaches 88.14%on the task of positive and negative emotion classification.(3)The dissertation proposed a cross-subject emotion recognition framework based on metatransfer learning and multi-scale residual network,which aims to solve the problem of large individual differences across subject.The framework reaches an average accuracy of 77.73%for emotion classification of 14 subjects on the self-built dataset,which is higher than traditional machine learning methods;in the DEAP dataset,the framework reaches an accuracy of 71.29%on valence task and 71.92%on arousal task,which is higher than the comparison method and proves the effectiveness of the framework.In summary,the dissertation designed an EEG monitoring system based on a novel flexible fabric.Under the induction of virtual reality scenes,an EEG data set that can be used for emotion classification was established,and a new cross-subject emotion recognition framework was proposed,which provides a new idea for reducing the influence of EEG individual differences in emotion recognition.