Research On Mental Workload Detection Algorithm

With the development of science and technology in modern society,some tasks have a high operational complexity such as manned aerospace,aircraft piloting,and submarine operation.It’s important for operators to evaluate and detect their mental load in real time.Controlling the working hours is of great significance to the physical and mental health and safety of the operators.Since the Electroencephalogram(EEG)features have high dimensions,the detection model is prone to overfitting.It is necessary to perform feature selection before training model.At the same time,feature selection is useful for researching the effect of different rhythms and electrode points in different brain regions on mental load.In the current research,the main feature selection methods of EEG features are principal component analysis and genetic algorithms.However,principal component analysis is an unsupervised dimensionality reduction method.It cannot measure the impact of each feature on mental load.Besides,the genetic algorithm takes a long time.Therefore,this paper proposes to use machine learning methods for feature selection.In addition,as mental load classifiers,traditional support vector machines and logistic regression algorithms are lacking in the study of combined feature weights.A factorization machine(FM)algorithm is proposed as a classifier to train EEG feature.This paper focuses on the mental load detection of simulated pilot.Fifteen participants were tested on low-wake,moderate-load,and high-load simulated flight missions,and EEG signals were collected under three types of missions.Firstly,the original EEG signal is removed artifacts.Secondly,energy feature of the signal are extracted by calculating the power spectral density.Finally,120-dimensional EEG features are obtained.Performing the methods of L1 regularization and Gini Impurity to select 120-dimensional EEG features.It is proposed that the electrode feature of the temporal lobe and occipital lobe regions under the?rhythm have an important effect on mental load.In order to achieve dimensionality reduction,the features in other regions can be filtered out.In addition,we hope to find an algorithm that can further explore the combined features due to the connection between the electrode points.This paper proposes to introduce Factorization Machine(FM)algorithm to perform the detection of mental load.This algorithm can improve the expression ability of the combined feature weights.Compared with the model without feature selection and traditional support vector machine algorithm,the mental load detection model combined with feature selection and FM algorithm proposed in this paper is more interpretable.Besides,the model generalization ability is improved.