Design Of Pulse Analysis System Based On Convolutional Auto-encoder Network

The pulse wave is a physiological signal produced by the heart’s regular beats,which is closely related to cardiovascular disease.By analyzing the shape,amplitude and rhythm of the pulse wave,it can be used for early prevention and auxiliary diagnosis of cardiovascular diseases.In traditional Chinese medicine,doctors use their fingers to feel the pulse of Cun Guan Chi,combined with observation,inquiry,auscultation and olfaction for diagnosis and treatment.This method of diagnosis usually requires long-term learning and accumulation of experience,and the results are related to the personal abilities of TCM physicians.Pulse wave signals are collected non-invasively through sensor technology,combining the disciplines of pulse diagnosis of TCM,electronic information,computer technology,pulse wave characteristics are extracted,and finally realizing quantitative description,repeated verification and intelligent identification of pulse waves,which is useful for the development and promotion of pulse diagnosis.It has theoretical significance and practical value.This research focuses on the key technical issues of the pulse analysis system,with the purpose of realizing pulse wave acquisition and analysis,taking the convolutional self-encoding deep learning network as the entry point,The pulse analysis system is constructed using electronic information technology and pattern recognition technology,finally,the pulse wave collection terminal、 host computer interaction system、 pulse data preprocessing and classification model design were completed.The main research work includes:(1)In order to realize the repeatability of pulse wave detection,the system requirements are analyzed based on the characteristics of the pulse wave signal,and a pressure-type pulse sensor is selected to build the pulse wave collection terminal.The pre-amplification circuit,band-pass filter circuit,double T notch circuit and voltage boost circuit are designed to condition the pulse wave signal,which helps to achieve stable acquisition of pulse wave signals.(2)For the analysis and management of pulse data,the overall framework of the pulse analysis system is designed based on the analysis of system requirements.The human-computer interaction interface of the pulse analysis system has been developed to realize the corresponding functional modules.Combined with database technology,the functions of visualization,analysis,and data management of pulse waves have been realized,and personal health files have been established.The function test shows that each module realizes the design function.(3)High-quality pulse data is the basis for subsequent analysis.In order to improve the system’s ability to analyze and process pulse waves,the pulse wave data is preprocessed.In the wavelet threshold denoising,the wavelet base sym8 is selected to perform 4-layer wavelet decomposition on the pulse wave signal,and the denoising effect of different threshold functions and threshold rules is measured by the signal-tonoise ratio.Secondly,the main wave peaks and troughs are extracted,through the dynamic differential threshold method and sliding window,and the trough is used to complete the period division.Finally,the length of the single-period pulse wave signal is fixed to 248 through period standardization,which is convenient for model input.(4)In order to solve the problem of limited labeling data caused by the high cost of labeling clinical pulse wave data,a pulse wave classification model based on the convolutional autoencoder network(CAE-Net)was constructed by reusing the pretraining model,which reduces the dependence on labeled pulse wave data.First,the self-learning of pulse wave features is realized based on the compression and reconstruction characteristics of the convolutional autoencoder(CAE),and the time domain feature point constraints are introduced in the reconstruction error to improve the CAE feature extraction ability.Second,CAE-Net is constructed by reusing the encoder’s structure and weights of the pre-trained CAE,and using labeled pulse wave data samples for network fine-tuning.Experiments on the collected data show that CAE-Net has an accuracy of 98% in identifying cardiovascular diseases,which meets the needs of the pulse wave analysis system.