Research On Respiration Rate Extraction Methods Based On Time-varying Features Of Pulse Waves In Time-frequency Domain

Respiratory rate is one of important physiological parameters of a human body,and it is also an important basis for clinical diagnosis of diseases.The promotion of daily monitoring of respiratory rate is of great significance to the prevention and control of respiratory diseases and other concurrent diseases.Most of the existing respiratory rate monitoring equipment is not suitable for daily monitoring due to the shortcomings of poor portability and complex operation,while the pulse wave collection equipment is light and easy to operate,and the pulse wave contains a variety of human physiological information,which is conducive to the daily monitoring of the respiratory rate.Although the method of extracting respiratory rate based on pulse waves has developed in recent years,pulse waves are very susceptible to interference and extracting stable respiratory information from pulse waves still has certain challenges,especially the performance based on big data is not satisfactory.Aiming at the above problems,this paper proposes a method for extracting respiratory rate based on multi-feature fusion of pulse waves.The main research work and contributions are as follows:(1)In terms of data preprocessing: This paper proposes a multi-channel data screening and fusion method.By analyzing and identifying the waveform characteristics of the pulse wave,the method realizes automatic detection of abnormal conditions such as distortion and sudden change caused by interference during the data acquisition process.Based on the detection result,the method can filter out abnormal pulse waves and retain high-quality pulse waves.In addition,this paper fuses the selected multi-channel pulse waves to make up for the lack of information in the single-channel pulse wave.(2)In terms of feature extraction and credibility evaluation: This paper proposes an extraction method for multi-dimensional time-frequency features of pulse waves and an evaluation method for feature credibility.In order to extract features containing respiratory information,this paper extracts multi-dimensional time-frequency features of pulse waves related to changes in amplitude and frequency from pulse waves after multi-channel fusion,which is based on the modulation effect of respiratory activity on pulse waves.Besides,in order to solve the problem that the quality of features is difficult to judge,this paper establishes credibility measures for each feature based on the periodicity and low frequency of respiratory activity.(3)In terms of model building: This paper proposes a method for extracting respiratory rate based on deep learning.First,this paper inputs the multi-dimensional time-frequency features of pulse waves into their respective convolutional neural networks to obtain multiple shallow features.Next,in order to improve the learning weight of high-quality features,this paper uses credibility measures to weight these shallow features and then input them into the deep neural network to extract the respiratory rate.The experimental results show that this method reduces the relative error by at least 3% compared with single feature,EMD decomposition and other respiratory rate extraction methods.