Mental Workload Classification Method Based On Domain Adaptation With Cross-session EEG Data

With the rapid development and widespread application of human-machine systems,operators working at high mental workload levels for long periods of time will quickly feel fatigued,leading to reduced work efficiency and increased probability of operational errors,resulting in safety accidents.Therefore,it is necessary to identify the mental workload status of personnel.Electroencephalogram(EEG)signals are commonly used for mental workload identification.Due to the non-stationarity of EEG signals,even for the same subject,there will be distribution differences in EEG signals collected at different times,specifically manifested as different marginal probability distributions and conditional probability distributions of data.In cross-time mental workload identification,as the data collection time interval increases,the generalization ability of the identification model trained by previous data will gradually decline.When the interval unit is days,the model’s identification ability is slightly higher than the random level.In order to improve the accuracy of cross-time mental workload identification,this paper studies a domain-adaptive cross-time mental workload identification method based on domain adaptation.This method includes three steps: The first step is to preprocess the original EEG signals and extract frequency domain features from the preprocessed data to obtain the subject’s EEG signal feature data.The second step is to use a domain adaptation algorithm to perform domain adaptive transformation on the source domain and target domain feature data to reduce the data distribution differences between them.The third step is to send the transformed source domain data and target domain data into traditional machine learning classifiers for training and identification respectively.The main work of this paper is as follows:(1)This paper maps mental workload feature data into a subspace and uses linear mapping to align the source domain and target domain in the subspace to reduce data distribution differences.A cross-time mental workload identification method based on subspace alignment is proposed.The experimental results show that the method has improved accuracy relative to the traditional mental workload recognition method using sliding window extraction features,with an average accuracy improvement of 7.03%.(2)This paper takes reducing the marginal probability distribution difference of EEG data as the learning goal of the model and proposes a mental workload identification method based on transfer component analysis.Experimental results show that this method has significantly improved recognition accuracy compared to traditional mental workload identification methods,with an average accuracy improvement of 11.52%.(3)This paper also takes into account the difference in conditional probability distribution of mental workload data,comprehensively considers the conditional probability distribution and marginal probability distribution of data,and proposes a cross-time mental workload recognition method based on balanced distribution adaptation.The final result shows that this method has higher recognition accuracy compared to traditional classification methods,with an average accuracy improvement of 17.64%.