Research On Sleep Stage Classification Method Of EEG Signals

Sleep stage classification is the division of several relatively stable states during sleep according to certain criteria.The EEG signal,which records the electrophysiological activity of the brain,shows different characteristics in the dissimilar sleep depth and possesses crucial research significance and parctial value in sleep staging.Accurate sleep staging can provide diagnostic basis for the treatment of sleep disorders,and then appropriate medical measures would be taken.In addition,potential sleep problems can be found from the alternation of sleep stages and sudden diseases,such as obstructive sleep apnea hypopnea syndrome,would be effectively prevented.Nowdays,because of the rapid development of modern society,sleep medicine research has become a hot topic with the increasing pressure of people’s life.Sleep staging by manual work is time-consuming and laborious,that is the reason why a large amount of research on automatic sleep staging have been done in recent years and this thesis is an experiment about automatic classification of sleep stage.After analysing the characteristics of EEG and the basic theory of sleep stages,this thesis can be divided into three parts: signal preprocessing,feature extraction and feature classification.The main contents of this thesis are as follows:(1)Analyze the characteristics of EEG and the basic theory of sleep stages.(2)Remove the interfering signal in EEG by wavelet transform.(3)In this thesis,the Hjorth parameter of time domain ferture,the frequency domain characteristic EEG rhythm wave energy and the nonlinear feature like Kc complexity,correlation dimension and approximate entropy are applied in feature extraction.Three commonly used methods of extracting rhythm wave,namely Hilbert Huang transform,wavelet packet decomposition and FIR bandpass filter are compared and analyzed in the process of extracting energy features.According to the comparison result,FIR bandpass filter is selected as the extraction method of rhythm wave.(4)An improved feature extraction method based on signal amplitude is proposed to overcome the aliasing of feature extraction results in N1 and REM phases.(5)The classification effects of traditional sleep stage classification method support vector machine and BP neural network are compared.The extreme learning machine is used as the classification method in sleep staging for the first time,and the effectiveness of the improved feature extraction method proposed in this thesis is verified.(6)The contents and experimental results of the thesis are summarized,and prospect the future work.