Research On Pulse Signal Recognition And Classification Method Of Human Health State

With the progress of society,industrialisation and urbanisation,the living environment and lifestyles of human beings have changed dramatically,and chronic non-communicable diseases have now become an important cause of death and disease burden for urban residents.According to the "China National Health and Nutrition Data" in recent years,the incidence of chronic diseases is increasing year by year,and the number of white-collar workers in a sub-healthy state is also on the rise.As a result,the monitoring of health status has become one of the most popular studies in modern society.The most common means of monitoring sub-health is still diagnostic scales,but these are inevitably influenced by the subjective thinking of the individual.To address the current shortcomings in the field of subhealth detection,the work done in this paper is as follows.A photoelectric volumetric pulse wave acquisition system was set up to obtain the experimental data for this study.A total of 73 school volunteers participated in this experiment,and each volunteer was asked to complete the SRSHS self-assessment form for pulse wave data acquisition,with each self-assessment form corresponding to their respective pulse wave data.Immediately afterwards,the noisy pulse wave signals collected during the experiment were processed using the CEEMDAN joint wavelet thresholding method.To address the problem of incomplete extraction of PPG feature quantities by current researchers,this paper extracts a total of 22 feature quantities from four directions,immediate domain,frequency domain,combined time-frequency domain and non-linearity.In the time domain direction,the ratio of each feature point on the pulse diagram and the area of the pulse diagram are extracted and calculated as feature quantities;in the frequency domain,the harmonic spectrum analysis and cepstrum analysis are mainly performed on the PPG signal,and the ratio of each harmonic to the normalized amplitude is extracted from its spectrum diagram as feature quantities;in the time-frequency analysis,the VMD decomposition is firstly performed on the PPG signal,but due to the selection of the modal number K and the penalty factor ? in the VMD However,the inaccurate selection of the modal number K and penalty factor a in VMD will have an impact on the signal decomposition results,so the grey wolf optimisation algorithm is introduced to optimise the VMD algorithm,and the optimised VMD algorithm is used to decompose the PPG signal and obtain the K eigenmodal components IMF,the Hilbert transform is performed on the obtained K components to find the marginal spectrum,and the SER in each frequency band is calculated as the time-frequency characteristic quantity;in particular,two In particular,two non-linear eigenvolumes,RCMDE and RCMFE,are introduced and their effectiveness is demonstrated experimentally.The 22 extracted features are identified and classified.In this paper,the RF algorithm is introduced to rank the importance of the feature set,and the extracted features are first filtered once to arrive at the most important 12 features.The traditional SVM classifier is also improved.In order to address the problem that most of the SVM classifiers improved by intelligent algorithms are easily trapped in local optima,this paper adds an improved grid search algorithm based on the PSO-optimised SVM,and the final recognition accuracy obtained is 97.50%.In addition,the accuracy obtained by the recognition and classification algorithm in this paper is compared with the accuracy of the traditional KNN algorithm and the classification accuracy in the references,which further proves the superiority and effectiveness of the detection and recognition method in this paper.