The Sleep Staging Methods Based On Feature Extraction Of EEG

As sleep involves in physiology, neurobiology, rehabilitation medicine, psychology, medical science and computer science, recently many countries have strengthen the study of sleep. Researches on sleep staging is not only the basis of diagnosing sleep related diseases, but the precondition of sleep quality evaluation, and has vital significance. Sleep staging work has long been artificial completed by sleep experts, but the staging rules between different experts are often difficult to obtain consistence. Automatic sleep staging gradually becomes a hot research area in late 80 s, but the recognition rate of current automatic sleep staging system is not high. Sleep stage is classified by every 30 seconds, so in traditional automatic sleep staging method, the features are extracted from every 30 seconds. Thus, we consider extracting features from different time slices and designing a system with high efficiency and accuracy.This paper firstly discusses the current situation of domestic and foreign study of sleep staging and the importance use of sleep EEG signal in sleep staging. Secondly it applies the wavelet packet coefficient, permutation entropy, Petrosian fractal dimension and random forests, support vector machine(SVM) in sleep staging. The innovations involved are below:(1) According to the characteristics of sleep EEG, we extract features of mean and standard deviation of the wavelet packet coefficient, permutation entropy, Petrosian fractal dimension from the sleep EEG by 30 seconds, 90 second, 150 second and 210 seconds. Through comparison of the three features in different sleep period, we find each feature shows regular changes in different sleep stage, so we use the three features as feature indexes for automatic sleep staging.(2) Introducing the random forests and support vector machine(SVM), we put the single feature of different time slices as an input to find the optimal combined parameters of classifier to explore staging system with highest accuracy. Ultimately, we determine that the sleep staging system of 210 s time slice of wavelet packet coefficiens, 30 seconds time slice of permutation entropy and 90 s time slice of Petrosian fractal dimension with SVM has higher accuracy.