Research On Sleep Monitoring Based On EEG Signals

Sleeping is one of the most important physiological activities for human body.The quality of sleeping makes great importance to people's lives and work.As the biological signal which reflects the activities of brain directly,electroencephalogram(EEG)has great concern with the quality of sleeping.There are many rhythm waves from different frequencies which can be used for sleep staging.In addition,the researches for now have shown that the number,form and density of sleep spindle have to do with sleeping quality and many sleeping diseases.So our project puts forward two algorithms in order to monitor people's sleeping quality,which are the extraction algorithm of sleeping EEG and the sleep spindle detection algorithm.Spindle is the typical characteristic of N2 in sleep staging.Thus the accuracy of spindle detection will be very high when you deal with the N2 EEG signal.However,it is very complicated to detect N2 EEG signal.Consequently,we propose we firstly detect the sleeping stage from EEG signal,and then design a spindle detection algorithm on sleeping EEG signal,which could balance the complexity and accuracy.In the extraction algorithm of sleeping EEG,we firstly design FIR low-pass filter to denoise for raw EEG signal.Then we can extract the different rhythm waves from EEG using many FIR band-pass filters.For purpose of finding out the effective feature to distinguish sleeping stage and awake stage,we utilize the tool of Principal Component Analysis(PCA).Meanwhile,we come up with the smoothing algorithms to deal with the problem of outliers.The principle of the adaptive algorithms is that the new feature variable is decided by the feature variable from the moment before and the new one in a different power respectively.Then,we take every 10 second of the EEG signal as a sample,with each sample labeled as sleeping state or awake state.Then the extraction of sleeping EEG is done by machine learning tool SVM.Finally,we use the new sensor TGAM to collect EEG signal from some experimenters and do some relative experiments.The results of the experiments prove the validity of our algorithmIn the sleep spindle detection algorithm,we choose the wavelet transform(WT)based on Mexican hat wave so as to obtain the energy feature of spindle more effectively.The probability of having a spindle at a given sample is defined as the proportion of the power spectrum of spindle and the total power spectrum.Traditionally,many algorithms always simply binarize the power spectrum of EEG signal according to their rankings,which will result in the loss of some significant spectrum information.We propose that different spectrum should be calculate into the probability in a proper power using Gaussian function so that the probability can represent the true situation.Besides,in order to estimate the probability more accurately,we use a sliding window smoothing algorithm to consider the relationship between each sample point and other information around it.In the end,we employ SVM classifier to detect sleep spindle as well.Because the traditional method to assess spindle detection algorithm is not considerate and not rational,we propose a more scientific assess algorithm.Also,we compare many other spindle detection algorithms and ours on the public DREAMS sleep spindle database.The experiment results show that our spindle detection algorithm which is based on WT and SVM behave better than other algorithms.