Research On Cross-time Mental Workload Detection With Multiple Physiological Signals

Mental workload refers to the cognitive resources that are payed to perform a specific task within a certain period of time.Studies have shown that mental overload and underload can lead to human error.By measuring the operator's mental workload and adjusting it in time,the task execution efficiency and production safety can be greatly improved.However,in the research of mental workload identification,there are still problems such as low correct rate and poor time robustness.This study intends to find the mental load characteristics by collecting four physiological signals,and use the mixed features to train the mental load classification model.Finally,a cross-time mental load identification method is proposed.In this study,a total of 20 healthy participants participated in the study.Each participant was required to complete two experiments with an interval of more than 48 hours.Each experiment required the subjects to perform mental workload tasks of different difficulty and collect electroencephalography,functional near-infrared spectroscopy,eye movement and electrocardiogram.The mental load sensitivity characteristics were extracted,and the mental load classification model was constructed.Some strategies were used to improve the time robustness of the mental load classification model.In the stage of feature extraction,the study concluded that the power spectral density of the partial frequency bands of the occipital region,the central region and the prefrontal lobe can effectively distinguish the mental workload state,and the temporal resolution is high;The pupil diameter,duration of the saccade,amplitude of the saccade,and the average saccade acceleration can distinguish mental workload.Relative pupil diameter is sensitive to changes in mental workload,and the results are consistent across differnent experiments;oxyhemoglobin and deoxyhemoglobin concentration can accurately distinguish different mental load states,but need to long sampling time;Heart rate variability was analyzed in time domain,and the average interval between normal R peaks can distinguish between different mental workload.In the modeling and classification stage,this study selected logistic regression,random forest,linear support vector classification and Nu support vector classification method to construct the model,and used single experimental data and cross-time experimental data for training.Finally,the linear support vector classification algorithm is selected to establish the mental load classification model.The input features are 98-dimensional EEG features and 2-dimensional eye movement features.In addition to improving the classification accuracy,the mental workload can be accurately classified using a shorter sampling time(1s)and a smaller feature dimension(100 dimensions).The highest classification accuracy rates of within-time model and cross-time model are 94.25% and 85.51%.This study obtained more mental workload sensitivity characteristics from a variety of physiological signals,and improved the correct rate of cross-time mental workload classification,and explored the field of mental workload measurement in depth,providing an algorithm for the development of real-time mental workload measurement equipment.