Mental Workload Detection Based On Multiple Bioelectrical Signals

In complex man-machine systems,operators need to make high-level decisions and take charge of supervision and monitoring,which leads to long-term mental concentration and high-pressure states.Excessive mental workload can easily cause mental fatigue and decreased alertness,negative burnout of operators could result in waste of human resources on the other hand.It is of great theoretical and applicable significance to maintain the monitoring of the mental workload of the experimenters for maintaining the safe and efficient operation of the man-machine system.Physiological indexes evaluation based on mental workload is currently recognized as a mental workload evaluation method.Traditional mental load identification methods rely on a single bioelectric signal,and the recognition rate needs to be improved.Comprehensive evaluation of mental workload by multi-bioelectrical signals has gradually become a research hotspot.In addition,most of the research on mental workload classification was based on traditional machine learning methods currently.In practice,the data distribution of mental workload will change over time,and the distribution is different among various subjects.The classifier trained from the first day's data could result in a large detection error when applied to the next day's data.A mental workload classification and recognition method based on transfer learning and integration of electroencephalogram(EEG)and electrocardiograph(ECG)was proposed to solve the low recognition rate of existing mental workload classification methods and the deviation of test samples.Based on the MATB-? platform of multi-task aviation situational operation,EEG and ECG signals of 12 healthy subjects were collected synchronously and preprocessed.The physiological information features were extracted from time domain and frequency domain respectively,and the mental workload detection was studied.On the basis of conventional mental workload classification methods,in order to synthesize the recognition results of different physiological information and avoid unsatisfactory final results caused by the low recognition rate of a certain physiological information.This paper proposed a decision-level weighted fusion method,which trained the classification models of EEG and ECG signals respectively.The classification decision function was extracted,the weight was calculated,and then the decision level is weighted and fused,the mental workload grades were output finally.Compared to the conventional classification of mental workload,this method could lead to better recognition accuracy when applied to non-cross-session mental workload detection.In terms of cross-session,transfer learning was introduced,and mental workload was classified based on Transfer Component Analysis(TCA)algorithm.Aiming at the difference of probability distribution between bioelectricity data about same subjects cross-session,bioelectricity data at different time was taken as the source domain and the target domain respectively,the source domain and the target domain were mapped into a common high-dimensional regenerated Kernel Hilbert Space,and the maximum mean discrepancy method was used to measure the distance.The edge distribution adaptation of the source domain and the target domain was realized through feature space transformation and feature dimension optimization.Compared with non-cross-session,non-transfer-conventional method,non-transfer-decision-level weighted fusion method and TCA method on cross-session issues,the experimental results showed that transfer learning could achieve higher recognition accuracy,which was 10.11% higher than that of non-transfer.Moreover,the recognition rate of mental workload based on multi-bioelectricity signals were higher than that of single bioelectricity signal.