Study On Drowsiness Detection Method Of Single-channel Electroencephalogram Signals

With the gradual acceleration of the pace of modern life,there are more and more problems caused by drowsiness.In a state of drowsiness,the brain will have problems such as reduced alertness,unresponsiveness and error-prone.This will affect our normal life and even cause great harm to the safety of our lives and property.At present,in the study of drowsiness detection,EEG signals are one of the most commonly used physiological indicators,because some characteristics of EEG signals will change as the human body becomes drowsy.Researchers at home and abroad have conducted in-depth research on the analysis method of EEG signals,and obtained relevant research results,which laid the foundation for the study of drowsiness detection of EEG signals.In this paper,the EEG data of 20 healthy subjects in MIT’s Sleep EDF database is used as the research object of drowsiness detection.According to the characteristics of EEG signals before and after drowsiness,the drowsiness detection methods of EEG signals were studied.The EEG signals were subjected to feature extraction and classification recognition,and some effective drowsiness detection methods were proposed to improve and develop new technologies for drowsiness detection.The main work of this paper is as follows:1.Power spectrum analysis of traditional rhythm electroencephalogram signals for drowsiness detection was performed.According to the labeling of EEG data in the Sleep EDF database by sleep experts,the original EEG signals are decomposed using haar wavelet packets transform to extract EEG signals in 5 traditional rhythms a,β and γ),and 3 new indicators based on power spectrum are proposed.The CPU time required for drowsiness detection by different indicators is analyzed.Compared with the existing 4 indicators,the results show that the three new indicators proposed in this paper have more significant differences before and after the wake-sleep transition time.And the newly proposed method has faster calculation speed.2.Variance feature analysis of traditional rhythm electroencephalogram signals for drowsiness detection was performed.Haar wavelet packet decomposition was performed on the original EEG signals,and the EEG signals of 5 traditional rhythms were reconstructed using wavelet packet coefficient sets,and then the variance of the EEG signals of 5 traditional rhythms in each cycle was calculated.And differences in the variance of EEG signals of 5 traditional rhythms before and after the awake-sleep transition time were analyzed.Finally,based on the analysis results,three new indicators for drowsiness detection were proposed based on the variance characteristics of traditional rhythm EEG signals.The results show that the drowsiness detection effect of the three new indicators proposed in this paper is better than the traditional rhythm,and the accuracy rate can reach 95.65%.Moreover,in the alertness detection,the three new indicators proposed in this paper also have good detection effects.3.An improved genetic algorithm based on support vector machine is proposed to calculate the optimal frequency band for drowsiness detection.Firstly,the support vector machine is used to identify the drowsiness of the subject,then the improved genetic algorithm is used to calculate the frequency band with the highest classification accuracy,and finally the leave-one-subject-out cross-validation is used to evaluate the classification performance of the classifier.The results show that the y rhythm has the best detection efficiency in the five traditional rhythms,and the accuracy rate is 80.94%.The detection accuracy of the new rhythm Rhythm(Ⅲ)(43.750000--48.046875 Hz)proposed in this paper is 89.52%.The detection effect of the new frequency band is better than the traditional rhythms.