Research On Automatic Staging Of Sleep Based On EEG Signal

Sleep is one of the most important physiological activities and behaviors of people.And it is the key factor in determining the quality of life.An effective and sufficient sleep has the function of restoring energy and physical strength and is a basic need to maintain health.In the environment of fast paced life and high-intensity work,sleep disorder has gradually becoming a kind of publicly harmful disease,which is common among ordinary people,the elderly and college students and has causing people to pay more intention to it.It is an effective method to objectively evaluate sleep quality and contribute to the study of sleep and sleep related diseases through understanding change of sleep status and storing sleep.There are four stages to achieve automatic sleep staging,including signal processing,feature extracting,feature selection and classification identification.Taking EEG signal as research object,a new method with higher classification accuracy is proposed.This method improves the feature selection stage and the classification identification stage.In the stage of feature selection,the hybrid feature method is adopted to extract features more effectively,and in the classification identification,Swarm Intelligent Algorithm is used to search parameters more effectively.The work done in each stage is as follow:1.Signal processing.The EEG signal is week and random.And it is very easy to disturbed by other signals,so the signal needs to be pre-processed to eliminate noise.Because of the characteristics of wavelet transform,it is widely used to eliminate noise.Wavelet threshold denoising method is used.Based on the research of other scholars,the wavelet basis function,the number of decomposition layers and the threshold value of wavelet transform can be determined to achieve the denoising preprocessing of EEG signals.2.Feature extraction.In the stage of Feature extraction,30 features are extracted from each sleep stage(30s is a unit)in four aspects.The number of time domain features is 11,which is maximum,minimum,variance,kurtosis,skewness,standard,averages,median,zero crossing value and Hjorth parameters.The number of frequency domain features is 3,including the value of center frequency,bandwidth and center frequency.Another aspect is time-frequency domain features,6 features include the ratio of γ wave,β wave,α wave,θ wave,δ wave and K complex wave occupies total energy.The last aspect is nonlinear dynamic features,10 features include c0 complexity,sample entropy,multi-scale entropy,fuzzy entropy,approximate entropy and singular spectral eigenvalues.After 30 features are extracted,the normalization process is performed in order to prevent the influence of the dimension between different features.3.Feature selection.In the stage of feature selection,a hybrid feature selection method is adopted.Firstly,decision tree algorithm and recursive feature elimination algorithm are used to select features.Secondly,based on the results of two feature selection methods,merging strategy is used to mix the results to obtain the input vectors of the final classifier.4.Classification identification.In the stage of classification identification,based on the actual sleep EEG data of MIT-BIT database,the improved fly fruit algorithm(IFOA)is adopted to optimize the penalty coefficient and kernel function parameters of support vectore machine(SVM)firstly,then the input features are trained by SVM to build a sleep staging model and the model is tested.The experimental results show that that the hybrid feature selection method can select more effective features than the single feature selection method.Moreover,compared with fly fruit algorithm,the improved fly fruit algorithm has strong optimization ability.Therefore,the automatic sleep staging method based on mixed selection and IFOA-SVM is an effective method,which provides a new idea for sleep staging.