Research On Sleep Staging Algorithms Based On Respiratory Signals

Sleep is an important part of human life.Good sleep can promote the recovery of people's mental state,attention,emotional control and judgment.However,with the accelerated pace of life,people's sleep quality is getting worse and worse,and staying up late,insomnia and other phenomena frequently occur.According to the survey,the proportion of adults with sleep disorders is as high as 30%.Therefore,how to make human sleep monitoring more simple and quick is of great significance for improving sleep quality and preventing sleep diseases.The traditional "gold standard" for sleep staging is polysomnography,but it has the disadvantages of high price,complicated operation,and invasive effects on the subject's sleep.At the same time,the rate of human respiration during sleep shows a rhythmic change similar to that of the brain,and the collection of respiratory signals is simpler than other signals such as EEG.The impact on human sleep is small,so this paper decided to use respiratory signals to study sleep staging.In this paper,the sleep breathing data preprocessing and feature extraction are firstly carried out,and then the improved random forest classification method for sleep staging is designed to realize automatic sleep staging.The main research contents are as follows:(1)Pretreatment of sleep breathing signal data.Because the sleeping environment is usually not absolutely quiet,it is often accompanied by human or environmental noise,and considering the intensity of the respiratory signal itself is weak,in order to filter out the high frequency noise in the respiratory wave,this paper designs the digital filtering by finding the optimal respiratory frequency.The device preprocesses the respiratory waves to optimize the data set.(2)Feature extraction of sleep breathing data.Feature extraction of respiratory data is performed from three analysis domains: time domain,frequency domain,and nonlinear.Firstly,the peak and valley of the respiratory signal are detected,and four time-domain characteristic parameters are extracted by calculating the number of respiratory peaks and the mean value of the respiratory peaks and valleys.The autoregressive model method is used to obtain the spectral energy of the corresponding frequency range of the signal,and obtain six Frequency domain characteristic parameters;two nonlinear features of respiratory data were extracted by approximate entropy and sample entropy.A total of 12 features were used as test inputs for the sleep automatic staging model.(3)Algorithm design and result analysis.Aiming at the randomness problem of random forest algorithm feature selection,this paper proposes two optimization strategies by introducing adaptive feature selection ideas: the first is multi-variable feature calculation method based on feature weight;the second is the feature of adaptive sparse constraint mechanism.The optimization method was used to optimize the random forest model training with the combination of normal and mild sleep disorders.Finally,the artificial AASM staging results and the SVM algorithm were used to verify the test output.The results showed that the accuracy of Wake-NREM-REM sleep staging was 83.7%,indicating that sleep conditions can be obtained based on the characteristics of respiratory data.In this paper,the sleep state is studied by the multiple characteristics of respiratory signals under different sleep states.The research results provide a method that can be applied to the automatic sleep staging system,which can be used as a supplement to the AASM rules and has a good development prospect.