Design Of A Sleep Monitoring System Based On Eeg Signals

In recent years,people’s lives have become increasingly stressful and the incidence of diseases associated with sleep problems has increased.Accurate sleep staging can provide a reliable basis for the diagnosis of sleep-related disorders.The EEG signal is an important indicator of the sleep-wake cycle and can be used as the main physiological signal for sleep staging studies.The monitoring of long-range sleep EEG signals is a prerequisite for automatic sleep staging.In this paper,we investigate the design of a sleep staging method and monitoring system for EEG signals,with the following main research work.（1）A study on the extraction of feature parameters of sleep EEG signals was conducted.In this paper,to address the problem that many sleep staging models have low classification accuracy and are prone to misclassification in the N1 and REM stages,a sleep EEG signal feature extraction algorithm based on symbolized amplitude difference to improve Permutation Entropy is proposed.The algorithm firstly symbolizes the amplitude difference of the sleep EEG signal and calculates the Permutation Entropy.Secondly,the mean value of the reconstructed subvectors is added as the weight to the permutation entropy calculation to obtain the symbolized amplitude difference Permutation Entropy,and a model for calculating the scale factora,which affecting the specificity of the algorithm is given.Then the sleep EEG data from Sleep-EDF sleep database is used for validation.The experimental results show that the EEG signals show significant specificity in the N1 and REM stages based on this algorithm,providing a new idea for the analysis of the heteronormative sleep associated with the N1 and REM stages.（2）Research on pattern recognition based on self-training least squares support vector machines.In this paper,to address the problem of slow training speed of standard support vector machines,a self-training least squares support vector machine with small sample training and lower computational complexity is proposed as the algorithm for pattern recognition in this paper.And the algorithm is tested for automatic sleep staging.The results are then compared with the automatic staging results of Bayesian linear classifier and artificial neural network algorithms.The tests show that the proposed algorithm in this paper has a combined index of 87.71%Acc and 83.06%Kappa,which is 3.24%and 2.45%more accurate than the other two algorithms respectively.In particular,for the more difficult to distinguish N1 stage,the Rec,Pre and₁ scores were 54.40%,58.97%and 55.51%,respectively,with a 9.05%improvement in₁scores over that of the staging model using energy features＆RNN.（3）Design of a portable EEG-based sleep monitoring system.This paper addresses the problems of expensive,bulky and uncomfortable wearing of commonly used multifunctional sleep recorders,and designs a sleep monitoring platform based on Mind Band Research EEG and acquires the sleep EEG signal from the Fp1 channel.The wavelet soft thresholding algorithm was then used to denoise the acquired real EEG data,followed by extraction of feature parameters and automatic sleep staging.The staging results show that the system agrees with the staging results of commercially available sleep monitoring bracelets by about89%,and can refine the light sleep phase of the monitoring bracelet statistics into N1 and N2 stages,which can achieve the basic function of sleep monitoring,and further validate the effectiveness of the algorithm proposed in this paper.