Mental Workload Identification Based On EEG Independent Components Features

Human-machine interactive operating system is more and more widely used with the development of Machine Learning,especially in high precision required and risk field,such as: Air Traffic Management System,the Aircraft Steering System and the Deep-Sea Stealth System.Operators need to have a good working state when performing tasks.In order to ensure the safety of personnel and the effective implementation of tasks,we can analyze their mental workload level to judge the mental workload state.Therefore,it is great meaningful to identify the mental workload levels of operators.The general method of mental workload identification is based on the EEG features.This method accuracy is low,because EEG is a mixed brain signal recorded from multi-channel electrodes(similar to multi-source mixed speech signal).It is unwise to directly analyze the mixed signals to distinguish the brain signal characteristics,so it is necessary to separate EEG signals into independent signal.ICA,a method to separate mixed signals,was first used in mixed speech separation.Despite ICA is widely used in EEG signal de-artifact,researchers use mixed brain signals without artifact to analysis mental workload.We use ICA separated the multi-source mixed speech signal to obtain pure signals for reference,and then we apply FastICA,the improved algorithm of ICA,to obtain independent EEG signals,in other words,EEG independent components.This paper proposes a mental workload identification method based on EEG independent components features for visual and operational tasks.This method can get better identification results because it uses independent components to extract features directly.For visual and operational tasks,we ask 10 subjects to conduct a multi-task simulation experiment,and collect different workload EEG signals under different level(low,medium,high)tasks.Firstly,we pre-process the collected EEG and use FastICA to obtain the EEG independent components.Then,we use the Power Spectrum Density(PSD)to calculate 120-dimensional EEG independent components features and introduce the information gain to obtain the best number of features.Finally,we applied Support Vector Machine(SVM)to classify them and proposed a method of mental workload identification based on EEG independent component features.We use the grid search to achieve the best parameter in training and introduce cross validation algorithm to avoid the problem of over fitting.Compared with the results of conventional EEG features and EEG independent component features,the average identification accuracy of the proposed method is improved by23.095%.In addition,the number of independent components in FastICA can be set manually,the experiment result revealed that the accuracy of mental workload identification is improved with the increase of the number of independent components.