Task-State Psychological Stress Detection Research Based On Single Channel EEG Signal

Modern society is full of stressors from work,family and health,which can easily lead to psychological stress.Continuous psychological stress if cannot be released in time,will have a negative impact on the physical and mental health of a person.Therefore,accurate detection,early detection and effective intervention and treatment of psychological stress are conducive to improving the physical and mental health of all people.However,at present,the detection of psychological stress is not satisfactory.On the one hand,it is because the clinical detection of psychological stress mainly relies on scales and doctors’ experience,which is highly subjective.On the other hand,it is because people are either unaware of their psychological stress status or are too shy to express it,which prevents them from detecting it in time.The psychological stress detection method based on human physiological signal analysis can effectively solve the above problems.Among them,the method based on electroencephalogram(EEG)signal analysis has the advantages of high specificity and non-invasive and non-destructive,which is expected to be an effective means to detect psychological stress objectively and quantitatively.Previous studies have shown that acquiring multi-channel EEG signals and extracting a large number of features containing asymmetries characterizing brain activity can detect psychological stress more accurately,which verifies the feasibility of EEG signal analysis in psychological stress detection.However,the acquisition of multi-channel EEG signals is relatively complicated,with high individual discomfort,poor wearability and high technical cost,which is not conducive to large-scale application.This study aims to explore the beneficial effects of single-channel EEG signals on psychological stress detection,and to provide effective methods and technical support for accurate detection and early detection of psychological stress based on wearable EEG.To this end,this study addresses the issues of single-channel EEG signal acquisition and validation,the separation of electrooculographic(EOG)components and its effect on detection results,and the mining of single-channel EEG information related to psychological stress.Based on this,a psychological stress detection model based on the joint analysis of multi-frequency band and multi-domain features of single-channel EEG signals was constructed.The main work and innovation of this study are as follows:(1)To address the problems of uncontrollable stressors and stress levels in psychological stress identification studies,four types of psychological stress induction experiments supervised by the scales were designed and implemented.A variety of stressors such as time limits and social evaluation threat were introduced to further induce psychological stress in the subjects.The results of the scale showed that all four types of psychological stress induction tasks designed could effectively induce psychological stress in the subjects,and the results of the subjects’ self-assessment showed that the degree of psychological stress induced by different types of induction tasks was different.(2)The single-channel EEG signals of 21 subjects were collected under the resting state and the four psychological stress-inducing tasks by using the self-developed headband singlechannel EEG signal acquisition device to construct an EEG signal dataset for psychological stress detection research.The results of signal validation experiments showed that the EEG signals collected by the device were stable and reliable,providing a wearable technology and signal reserve for single-channel EEG signal-based psychological stress detection.(3)To investigate the influence of EOG components on the results of task-state psychological stress detection,an EOG component removal method based on a deep learning framework was developed.Based on publicly available data,a large sample size dataset was constructed,and a hybrid model of a one-dimensional convolutional neural network and a long and short-term memory network was constructed based on the respective advantages of convolutional neural networks and long and short-term memory networks for EEG analysis,which was trained and validated to achieve the desired de-eyeing effect.Through ablation experiments,it was demonstrated that the constructed hybrid model enhanced the sub-model performance.Based on this,the effects of removing the EOG components on the psychological stress detection results were compared,and it was found that the single-channel EEG signals based on both before and after the removal of the EOG components could achieve an average detection accuracy of more than 88%.The results were better before removing the EOG components.This confirms that for detecting task-state psychological stress,the EOG components of the EEG signal is valid information and should be retained.(4)In order to fully explore the effective information contained in the single-channel EEG signal that can characterize the psychological stress state,a joint analysis method of multifrequency band and multi-domain features of the single-channel EEG signal was developed.The proposed method extracted time domain,frequency domain,time-frequency domain and non-linear features from the five sub-frequency bands of EEG signals and the overall signal,and constructed the optimal feature set by three different feature selection methods,namely mutual information(MI),random forest(RF)and recursive feature elimination(RFE),and fed them into a support vector machine(SVM)to identify psychological stress.The accuracy of the proposed method in detecting four different types of psychological stress states was 93.1%,85.8%,88.9%and 85.8%,respectively,all higher than the optimal results of 90.5%,78.9%,85.6%and 83.9%for single domain features and 78.6%,76.9%,86.1%and 80.6%for single frequency band signals.This confirms that the joint analysis of multi-frequency band and multidomain features can provide more valid information related to psychological stress than singlefrequency band signals or single-domain features of single-channel EEG signals.In this study,we explored the feasibility of detecting psychological stress based on singlechannel EEG signals,and constructed a psychological stress detection model based on the joint analysis of multi-frequency band and multi-domain features of single-channel EEG signals,which provides a feasible method and technical support for early detection and intervention of psychological stress.