Research On Estimation And Improvement Of Human Sleep Quality Based On Deep Learning

Sleep is essential for people,and it is our main way to restore mental state and physical strength.However,with the acceleration of the pace of modern society and the increase of work and life pressure,more and more people suffer from sleep disorders such as insomnia.Not only have the people been tortured physically and psychologically,but also have been seriously affected in learning and working efficiency as well as the life quality.Traditional insomnia treatment methods only increase the amount of sleep to a certain extent,and do not consider from the perspective of sleep quality,so it cannot effectively alleviate various pains caused by insomnia.In response to this problem,a non-drug intervention therapy method based on sleep state identification has been proposed,which is on the basis of the correspondence between different sleep states and EEG signals,to design an efficient and accurate sleep state identification algorithm,and then to induce the body in stages with physical factors,so as to achieve continuous cycle control of the sleep process and improve sleep quality.Based on this starting point,the research work has been carried out in the following three areas:(1)Research on identification algorithms of sleep state.Traditional machine learning methods and deep learning methods are used to construct three different identification models to identify human sleep state,respectively.The identification model based on traditional machine learning methods takes two non-linear feature parameters as inputs,that is multi-scale entropy and power spectral entropy designed artificially.And the SVM classifier performs sleep state identification.While the identification model based on deep learning sets feature extraction and classification as a whole,which takes the STPS time-frequency spectrum as input,and uses CNN and ResNet to extract the features hidden in the EEG signal and complete the sleep state identification,respectively.And it can effectively avoid the uncertainty caused by artificial designed features.(2)Research on sleep process control methods.Based on the accurate identification of the human sleep state,the regulation effect of different physical factors on the sleep process of the brain are discussed according to the EEG biofeedback theory,and then a sleep control system based on sound is established.The sleep state of the brain is taken as the controlled object,sound is used as the control variable,the sleep state detection system is considered as the sensor,and the sleep system controller is applied on the human body according to certain control rules to achieve the precise control of the sleep state.(3)Evaluation of human sleep quality.According to the research ideas in this paper,a large number of human sleep experiments are performed.The Emotiv EPOC acquisition device is used to collect the EEG signal during the subject's sleep process in real-time,the sleep state identification method proposed is used to identify the subject's sleep state,and a network model with the best identification effect is selected.The output of the identification model is used as the feedback of the sleep control system.Subjects are induced under different sounds according to the established sleep control rules to realize the phase transition of the sleep state of the brain.The experimental results show that the sleep state identification algorithm used in this paper has a high identification accuracy of 93.5%for the six sleep states,and the sleep control system constructed according to the sleep control rules can improve the human sleep quality significantly.