Study On Sleep Feature Extraction And Staging Methods Based On EEG Signal

With the gradual acceleration of the rhythm of modern life,the pressure on people's lives continues to increase,and more and more diseases are caused by sleep problems.Therefore,cross-disciplinary sleep medicine has attracted widespread attention from medical research institutes and universities.At present,in the study of sleep clinical diseases,relying mainly on expert personally on-site analysis,due to objective reasons and artificial subjective misjudgment,etc.,resulting in the accurate of sleep staging is lower.In recent years,medical experts and scholars have devoted themselves to the study of sleep EEG automatic staging based on signal processing theory and pattern recognition algorithms.However,foreign progress has been relatively good,and domestic lags behind.The use of sleep EEG to assess the accuracy of sleep quality needs to be further improved.The thesis firstly analyzes the characteristics and staging criteria of sleep EEG signals,and studies the sleep electroencephalogram monitoring data of 8 subjects in the MIT-BIH Sleep-EDF database based on EEG sleep feature extraction and staging.As the object of sleep analysis,the features of EEG signals are extracted:wavelet transform algorithm is used to extract energy features of rhythm waves;multiscale entropy algorithm is used to extract entropy features of EEG signals at different scales;then principal component analysis(PCA)method is used combines and reduces the energy of the rhythm wave and the entropy of different scales to simplify the model structure of the classifier and reduce the training time.In the sleep feature staging process,the pattern recognition method was used to analyze the sleep period.The BP neural network and the SVM classifier were used to establish the stage model of sleep characteristics.The four stages of Wake,SWS,LS,and REM brain electrical signal characteristics were classified to verify the automatic sleep.The effectiveness of the staging system.The experimental results show that the wavelet energy and multi-scale entropy proposed in this paper are the characteristics of the sleep phase,which can effectively reflect the time-frequency and nonlinear characteristics of the sleep period.The principal component analysis method can reduce the original redundant features while retaining the original sleep characteristics of most EEGs,BP neural network and S VM as a stage model of the sleep stage show good classification and classification effects on EEG sleep characteristics.Therefore,this paper proposes a sleep-feature extraction and staging method based on EEG signals that can achieve high accuracy and stability,providing a new exploration for sleep medicine research.