Research On The Pulse Signal Processing Method Based On Wavelet Transform

Pulse diagnosis is an ancient and magical medical skill, development so far, ithas formed a large and complex theoretical system, however, compared with therapid development of high technology, the process of the digitization of pulsediagnosis develops slowly, it is not able to combine with modern advancedtechnology well, leading to it developed slowly, and it is difficult to spread, theresearch on the digitization of pulse diagnosis is imperative.In this article, the study included pulse signal acquisition, preprocessing,feature extraction, classification, gradual process, expect to be able to contribute tothe research on the digitization of pulse diagnosis. Mainly to complete the work asfollows:Pulse signal acquisition: samples were from the various departments ofHospital of Guangdong Province, including: department of physicalexamination(101cases of healthy samples), Nephrology(136cases of chronickidney disease patient samples), Endocrinology(134cases of diabetes patientsamples), the sampling device is a composite of pressure and pulse photoelectricsensor acquisition device.Pulse signal preprocessing includes: using wavelet decomposition and singlebranch reconstruction to remove noise, using characteristic on single-period of pulsesignal to cut off inactive ingredients, and the method to remove baseline drift isextract every start point of one period, do three spline interpolation on the startsequence to simulate baseline drift, and then subtract it, we can achieve the purposethat to remove the baseline drift, about period segmentation, by comparing the therising edge length, falling edge length of one period to determine the startsequenceand the shape of pulse signal, then we can get every period signal. Finally,to normalize the amplitude of each period signal to the range of0to1.Feature extraction:we extracted12-dimensional time-domain characteristics inthis article, including: the number of peaks of pulse signal, the length of time fromstarting point to the main peak; the period length of pulse signal and so on.Meanwhile, using dmey wavelet to decopose pulse signal into five levels, we can get6-dimensional wavelet domain features.Classification: Based on the above features using k-nearest neighbor classifierand support vector machine for classification experiments. The experimental resultsshowed that: the distinction between health and chronic kidney disease onenergyfeatures, support vector machine is better; the distinction between health and diabetic on time domain, support vector machine is better. Finally, classification onthe combined characteristics and health and disease samples, overall, the recognitionrate increased and the feasibility of this research has been verified.