Research On Non-contact Human Respiratory Detection Method Based On WiFi-CSI

As one of the important physiological activities,respiration reflects the health status of human body.In China,chronic diseases,including chronic respiratory diseases,account for more than 80% of the deaths in the country.In the context of the outbreak of COVID-19,the normalization of breath detection is particularly important.Therefore,breath detection technology has a wide range of applications in the field of intelligent medical care,home care and sleep detection.With the development of wireless technology,non-contact breath detection based on Wi-Fi wireless sensing has been widely studied because of its convenience,safety and easy deployment.However,the current research on breath detection based on Wi-Fi still faces three main problems:(1)Lack of a general and accurate human perception model.(2)The estimation error of respiratory rate is large,which cannot meet the strict commercial medical needs.(3)The classification of respiratory patterns is not clear and detailed,and there is a lack of detailed classification of respiratory patterns based on Wi-Fi.The research is carried out in view of the above problems:(1)In order to solve the problem of lack of reliable breathing detection model,a human breathing model based on Fresnel zone theory is established.On this basis,the relationship between the dynamic and static transmission path of RF signal and the received data in Fresnel zone environment is studied.The periodicity of the CSI respiration signal is revealed,and the optimal detection position is deduced.On the basis of the Fresnel zone detection model,the research of human breath detection based on Wi-Fi CSI is realized,and the feasibility of millimeter-level breath information detection by wireless sensing is answered,which provides specific model construction and theoretical guidance for this study.(2)Aiming at the problem of large error in the estimation of respiratory rate,the research on the estimation algorithm of respiratory rate is carried out.The algorithm flow includes data preprocessing,subcarrier selection,false peak removal and frequency estimation.The Hampel filter is used to filter out the abnormal values of the original signal,the improved wavelet threshold denoising filter based on the open method is used to effectively remove the noise,and the least square method is used to remove the DC component;The variance maximization is used for selecting the sub-carrier to reduce the calculation redundancy;The false peak elimination algorithm is used for effectively removing the jitter signal caused by the fluctuation of the human body;Finally,an accurate estimation of the respiratory rate with a minimum experimental error of 3.9% is achieved by the respiratory rate estimation based on the cross-translation points.(3)In order to solve the problem that the classification of breathing patterns is not clear and detailed,the breathing pattern classification algorithm is studied.Breathing is further divided into three modes: normal breathing,pause breathing and deep breathing.The data set is enhanced by curve fitting technology to solve the problem of scarcity of acquisition data.In addition,CSI data are encoded into two-dimensional images by Gram angular field method and sent to four frames of convolutional neural network for training.Meanwhile,combined with the characteristics of breathing patterns,a CSI breathing pattern classification algorithm based on BI-AT-GRU is proposed by further adding bidirectional and attention mechanisms to the gating cycle unit.The experimental results show that the classification accuracy of the convolutional neural network is 93.33%,the classification accuracy of the gated recurrent unit is 94.17%,the accuracy of BI-AT-GRU network classification is 96.67%,which verifies the superiority of the network in the classification of CSI breathing mode.The thesis contains 68 figures,15 tables and 84 references.