Research On Non-contact Sleep Disorder Sensing Technolog

Sleep,as a basic human need,is an essential foundation for a person to stay healthy.Sleep disorders and sleep stages are essential for assessing sleep quality and are a proven biometric technique in diagnosing cardiovascular and other diseases.Radar-based non-contact sleep monitoring combines comfort and convenience with guaranteed measurement accuracy and can be an effective alternative to existing sleep monitoring.In this thesis,we use non-contact bio-radar to monitor patients during sleep period,preprocess the raw signal with DC removal,phase extraction and trend term removal,extract the radar signal for the whole night,propose improved sleep staging based on viewable network and multicenter sleep staging algorithm based on domain adaptation based on the characteristics of multicenter data domain,realize the classification of four sleep stages by radar abnormal respiratory events are detected and their mapping relationships with sleep stages and related diseases are analyzed.The main work of this thesis is as follows:1.The FMCW radar-based signal acquisition system used in this thesis is introduced.Firstly,the hardware platform of the radar system and the corresponding signal processing methods are introduced;the obtained raw signals are subjected to DC removal,phase extraction and trend term removal to obtain accurate vital sign signals;finally,the sleep data set is built by combining the radar signals and the corresponding PSG results.2.Deep learning-based sleep staging was completed.Firstly,for the problem of unbalanced sleep stage data in sleep dataset,the processing of oversampling,undersampling,adding classification weights and Focal Loss was carried out,and the processing based on oversampling showed the best performance;then the convolutional neural network was constructed,and the sleep stage classification was completed with one-dimensional time-series signal as input,and the classification accuracy of 71.4% and Then the improved sleep staging method based on viewable network was proposed,introducing the concept of viewable network,mapping the one-dimensional vital signs temporal signal into a complex network,extracting the degree distribution of the network combined with the original temporal signal to construct a dual-input network architecture,which improved the accuracy of four classification to 73.6%and the Kappa coefficient to 0.48;finally,according to the data The data set was divided into different distributed data domains according to the characteristics of different devices and collection locations of the centralized control experiment,and the concept of domain adaptation was introduced to complete the multicenter sleep staging based on domain adaptation,which further improved the adaptability of the model among different distributed data domains.3.The mapping relationship between sleep disorders and sleep stages was analyzed.Firstly,the feasibility of radar monitoring abnormal breathing events was verified by comparing hypoventilation and apnea event compliance with PSG;an algorithm for detecting abnormal breathing events based on signal amplitude was investigated,and the experimental results showed that the correlation coefficient between radar and PSG apnea hypoventilation index detection results was 0.96;finally,the mapping relationship between sleep breathing disorder and sleep stages and abnormal breathing events was analyzed The relationship was analyzed,and the results showed that the severity of sleep apnea in patients with obstructive sleep apnea hypoventilation syndrome(OSAHS)worsens during the rapid eye movement phase,while there is an association between OSAHS and hypertension and cardiovascular diseases.