Research On Slow-wave Sleep Detection Algorithm Based On Breathing Audio

Sleep is an indispensable physiological phenomenon of human beings.Sleep accounts for nearly one third of the life of a person.The quality of sleep is closely related to human health.Slow-wave sleep(SWS)is the most important sleep stage in human sleep.Slow-wave sleep can help individuals recover from daily fatigue.The lack of slow-wave sleep may also cause some chronic diseases.The quality and quantity of slow-wave sleep can reflect the quality of a person's sleep,so an effective detection method of the slow-wave sleep is of great significance to people's daily life and work.A slow-wave sleep detection method based on breathing audio signal is proposed,which can effectively realize slow-wave sleep detection.This method records audio signals based on audio sensors,and then performs pre-processing such as framed windowing,noise reduction,and sound event detection on the audio signals to obtain respiratory audio event signals from the audio signals.By extracting the effective slow wave features in respiratory audio events to detect the slow-wave sleep stage in the whole sleep.Its main work is as follows:1.Framed windowing in audio signal preprocessing is introduced.The methods of spectral subtraction and Wiener filtering noise reduction are studied.Based on the characteristics of weak breathing signal and low signal-to-noise ratio,a sparse non-negative matrix factorization denoising method based on linear constraints is proposed.This method combines the fundamental frequency of the breathing audio signal to construct a breathing dictionary matrix to linearly constrain the non-negative matrix factorization.Compared with the traditional noise reduction method,it can better filter out low-frequency noise in a low signal-to-noise ratio environment.Based on the characteristics of strong periodicity of respiratory signals,a method for detecting sound events based on spectrum correlation is proposed.Compared with the traditional zero-energy-based sound event detection algorithm,the accuracy of detecting respiratory events increased from 78.5% to 87.7%.2.The traditional acoustic characteristics and frequency domain characteristics of respiratory audio signals are introduced.Based on this,combining the concept of signal variability,18 slow wave characteristics such as respiratory variation coefficient,respiratory consistency,and respiratory intensity were proposed to describe the characteristics of respiratory signals during slow-wave sleep.3.A slow-wave sleep detection method based on respiratory events is proposed.The method first performs preprocessing such as cleaning and normalization on the respiratory event features to obtain the preprocessed respiratory event feature set,and then performs a secondary screening on the respiratory event features to further filter out non-respiratory event features,then,in order to solve the problem of imbalance of slow wave sleep data,a balanced bagging classifier is used to classify the slow wave sleep stages of respiratory events.Finally,the breathing events of the same sleep segment are voted to determine the slow wave sleep classification results of the sleep segment.Compared with the traditional slow wave sleep detection method,the method in this paper can better reflect the difference of different respiratory events in the sleep segment.4.A sleep experiment platform was set up,and 23 audio breath data were recorded.The effects of slow wave characteristics and feature selection on the classification results of slow wave sleep detection methods were analyzed,and then the performance of the slow wave sleep detection method based on respiratory events proposed in this paper is compared with the traditional method.The experimental results show that the performance of this method is better,and the accuracy rate of 90.4% and kappa coefficient of 0.637 can be achieved.