The Fatigue Estimation And Real-Time Monitoring Based On Eeg

The 21st century has been considered the"brain science era". Aiming at discoveringthe secrets of the most complex part of human body, brain science has becomeone of the most attractive branches of nature science. Electroencephalogram (EEG) isthe electric potential on the scalp produced by the firing of neurons within the brain.EEG contains plenty of information related to brain state, including fatigue level. Analyzingand monitoring the fatigue level from EEG signals is the result of a successfulcombination of computer science and biology. It is also good opportunity for us toknow and utilize information contained in brain.Operators in many positions of jobs always face the problem of inevitable vigilancedecrement as they have to repeat monotonous operations for long periods of time.Automatic analysis and detection of fatigue level is an important issue to avoid the accidentin these operations. Based on the background and physiology foundations above,this paper studies fatigue estimation technology and proposes a complete frameworkfor processing EEG signals and estimating fatigue.In this paper, a fatigue estimation solution based on Sparse Representation Classification(SRC) is proposed. It firstly extracts frequency domain feature from EEGsignal, and then applies random dimension reduction method. Finally, SRC is used toestimate fatigue level. The result shows that this solution can recognize fatigue efficiently.It can use less features and training data to get 90% accuracy. The effect offeature selection method is very small to use this solution.The major work of this paper is as followings:Firstly, establish a set of complete solutions of collecting, processing EEG signals,and estimating fatigue level. The proposed technologies include: (a). EEG signal collection from simulated driving experiment and picture recognition experiment. (b).Artifact removal. (c). EEG labeling. The EEG segments are divided into alert, drowsyand sleep. (d). Feature extraction. EEG features are extracted from frequency domain.(e). Feature selection. (f). Fatigue estimation. SRC, SVM and hGMM are all usedto classify fatigue level. The procedures above can be used as a general solution for aEEG processing system or fatigue analysis system.Secondly, according to characteristics of fatigue shown on electrical frequencyspectrum, four kinds of frequency domain features are extracted, such as Power Proportion,Variance of Power, Average of Frequency and Variance of Frequency. 150 featuresare selected from over thousand features based on minimum-redundancy maximumrelevancymethod. The analysis of the importance features which reflect fatigue is alsogiven in this paper.Third, Sparse Representation Classification is used for brain fatigue estimation.Random dimension reduction is used to reduce EEG feature dimension. High accuracyfatigue prediction model is created using less training sample and less features.Finally, EGCIL, a kind of real-time fatigue estimation software is designed. Fatigueestimation methods mentioned in this paper are integrated in this software whichcan estimate fatigue level based on EEG under real-time and simulated environment.