Research On Abnormal Breathing Identification Technology Based On Vital Signs Radar

Respiration is one of the important physiological parameters of the human body and is closely related to the health of the human body.Monitoring respiratory signals can assist in the diagnosis and treatment of diseases.Traditional respiratory disease detection technology requires professionals to operate,is expensive,and can easily cause discomfort in contact with the human body,is not suitable for long-term monitoring.The non-contact respiratory signal monitoring system based on vital signs radar can overcome these drawbacks,and has the advantages of all-weather,high sensitivity and strong penetrating power.It is not easily affected by the environment and is more suitable for long-term monitoring of human respiratory signals.This paper uses vital signs radar to monitor human respiratory signals,extracts classification features for respiratory signals based on sparse representation algorithms,and realizes the classification of normal breathing,Cheyne-Stokes breathing,and dysrhythmic breathing through convolutional neural networks.Then Cheyne-Stokes breathing related features of cardiac function were analyzed.The main work of this paper is as follows:1.The respiratory signal monitoring system based on vital signs radar selected in this paper was introduced,and the hardware platform and the structure of the radar module were explained;One band-pass filter was designed to extract the respiratory signal from the radar echo signal.The radar echo signals characteristics of the three different breathing patterns were analyzed in detail.2.Aiming at the problem that the time domain and frequency domain features commonly used in the extraction of respiratory signals from clinical patients are difficult to achieve accurate classification of breathing patterns,a sparse representation-based feature extraction method for respiratory signal classification is proposed.Based on the sparsity of respiratory signals in time-frequency domain and wavelet domain respectively,by constructing Gabor dictionary and wavelet dictionary,the respiratory signals is projected to the corresponding sparse domain,then an orthogonal matching pursuit algorithm is used to obtain the sparse solution,and after determining the optimal sparsity of the sparse solution,the non-zero element items in the sparse solution are used as the sparse domain features of the respiratory signals.3.The algorithm flowchart of recognition and analysis of abnormal breathing was introduced.First the breathing pattern classification algorithm based on the convolutional neural network was researched,then the convolutional neural network model was constructed,using the feature data set to adjust the parameters and optimize the model to determine the best network structure.Since the feature parameters of Cheyne-Stokes breathing of patients with heart failure are related to cardiac function,the feature analysis algorithms based on short-term energy and empirical mode decomposition are respectively studied for the related features of Cheyne-Stokes breathing.4.The experiments were designed and the experimental results were analyzed.First,the clinical experiment of respiratory signal monitoring is designed,and the feature data set is set after data collection and analysis.Experimental data results show that the classification accuracy of the convolutional neural network breathing pattern classification algorithm based on sparse domain time-frequency features and wavelet features has reached 91.8% and 97.9%,respectively.Compared with the existing breathing pattern classification algorithm based on laboratory simulation breathing data extracting time domain and frequency domain features,the accuracy rate is increased by 5.6%.Secondly,according to the breathing difference between different people,the generalization performance test of the model was carried out.The experimental results showed that when predicting the breathing pattern of subjects who did not participate in the model training,the average classification accuracy of time-frequency features and wavelet features reached respectively 85.3% and 89.6%.Finally,the consistency evaluation of the feature analysis algorithm based on short-term energy and empirical mode decomposition adopts the Bland-Altman diagram.The results show that the respiration feature analysis algorithm based on empirical mode decomposition is more consistent.