Human Mental Fatigue State Recognition And Its Improving Methods

In recent years, mental fatigue has become a kind of disease which is prevalent inpopulations in modern society, not only people’s working and learning efficiency waseriously affected by long-term mental fatigue, but also the health and safety of human life,thus it is critical to be able to identify the mental fatigue correctly.In this paper, mental fatigue was identified through analyzing the EEG signal ofhuman life, and the improvement of mental fatigue was proposed. Five kinds ofexperimental condition was designed, they are state of awake, state of first fatigue, state ofrestore from resting with eyes closed , state of second fatigue, state of restore fromlistening to music. Subjective measurement experiment accomplished by the form offilling the sleep scale table, the result showed that, subjective evaluation scores rose underthe state of mental fatigue. EEG signal analysis is divided into the following steps: thefirst step, we preprocessed the EEG signal, including down sampling, lead selection andband-pass filter. The second step, the features of EEG signal are extracted, including theaverage relative energy of the four rhythm wave, the average wavelet packet entropy andthe coefficients of AR power spectrum estimating, and three types of feature vectors arecombined into a high dimensional feature vector. The third step, the state ofmental-fatigue and non-mental-fatigue was identified by the SVM classification, first theoptimal parameters was found by the grid search algorithm, and then cross validationmethod was used for classifier training, the accuracy rate of training classification is at98%; last test classification was conducted, the accuracy rate of test classification is at97%. The fourth step, the validity of two improvement for mental-fatigue was verified,firstly several commonly used improvement for mental-fatigue was put forward, and thenfour kind of classification samples were classified with the classifier, the accuracy rate ofclassification is at 96%, 92%, 90% and 85% respectively.EEG signal analysis is accomplished through the software of MATLAB7.0, the focusof analysis include: first, the verification for effectiveness and feasibility of the fatiguecharacteristics; second, the selection of wavelet function and decomposition level; third,the selection of order and algorithm for AR power spectrum estimation; fourth, the selection of optimal parameter for SVM classification.In this paper, the features that can effectively reflect the state of mental-fatigue wereextracted, the classifier which can correctly identify the state of mental-fatigue wascreated, and that we can improve the state of mental-fatigue by resting with eyes closedand listening to the music was validated, these provide reference for mental-fatigue onfollow-up study and are of great practical significance.