Research And Implementation Of Human Fatigue Realtime Monitoring Method Based On Physiological Signals

Human fatigue monitoring based on physiological signals refers to using physiological signals collected from human body to judge the fatigue degree.Fatigue monitoring based on physiological signals is very promising for practical application because of its objectivity and continuity,but the physiological signals of different users are different due to the differences in personality,age and other physiological and psychological aspects.The difference between users makes the fatigue monitoring model established on existing users become less stable when it is directly applied to new users.In this paper,two cross-subject mental fatigue monitoring algorithms based on Domain Adaption and Domain Generalization are proposed,and a fatigue monitoring framework for civil aviation pilots is proposed.The main work of this paper is as follows:1.Aiming at the problem that the existing mental fatigue monitoring algorithm based on domain adaptation technology is prone to produce boundary confusion,leading to the decrease of the accuracy of cross-subject monitoring model,an algorithm named Spatialtemporal Maximum Classifier Discrepancy(STMCD)is proposed based on spatio-temporal features and maximized classifier differences.By extracting the spatial temporal features of EEG signals and taking full advantage of the differences of the classifiers in the new user's data,the algorithm aligns the distribution of the new user's data with the existing data to reduce boundary confusion.Experimental results show that the accuracy of the proposed STMCD algorithm is improved by 8.17%,2.28% and 17.46% respectively on three public datasets.2.The existing mental fatigue monitoring algorithm based on domain generalization technology only considers the aligned edge distribution,which leads to the decrease of the accuracy of the cross-subject monitoring model.In this paper,an algorithm named Conditional Alignment Domain Alignment(CADG)is proposed.By allocating the edge distribution and conditional distribution of the EEG data of the existing users,the algorithm can extract the mental fatigue features irrelevant to the users,so as to reduce the influence of user differences in fatigue monitoring on the new users,and obtain better generalization performance for the new user data.Experimental results show that CADG algorithm improves the accuracy by 2.10%,2.12% and 2.59% respectively on three public datasets.3.Most of the existing human fatigue monitoring methods use EEG signals,while EEG acquisition equipment greatly interferes with pilot activities.Therefore,in this paper,ECG signals,which can be collected by convenient wearable devices,are used to monitor pilot fatigue.To alleviate individual differences of pilots' ECG signals,a framework named Evolutionary Pilot Mental Fatigue Monitoring(EPMFM)is proposed,which includes cold start,model calibration and model customization.Individual differences can be gradually reduced with the accumulation of pilots' ECG data.Experimental experiments on the collected ECG dataset of pilots show that the accuracies of three stages are increased by 8.01%,18.07% and 38.95%,respectively,compared with baseline.