Pulse Signal Features Analysis And Research Based On Ensemble Empirical Decomposition And Stationary Wavelet Transform

Pulse is one of important physiological features of human body, which is causedby cyclical contraction and diastole. It is a combination of heart, artery elasticity andblood information, such as volume, viscosity, and can be used as diagnostic indicatorsof cardiovascular diseases and its bad degree. In recent years, with the development ofsensor, measuring and saving of pulse signal have become very convenient. Now thepulse signal features analysis is a hotspot and difficulty in the signal processing field.Therefore, to realize automatic analysis and recognition of the pulse signal has thevital significance in the aspects of the noninvasive diagnosis and remote monitoring,etc.In this paper, the main research results and innovation points include thefollowing aspects:1. Aim at the problem of removing the noises in pulse signal; we used a theory ofempirical mode decomposition (EEMD) which is gradually developed in recent years.Compared with wavelet transformation, it has more adaptability, and signal featuresare more clearly. Compared with EMD, it can effectively solve the problem of modalaliasing, and the IMF components’ physical meaning is more obvious.(1) Forlow-frequency baseline wanders, a method of adaptive baseline drift correctionalgorithm based on the EEMD is proposed. On the premise of minimum of the signaldamage, we achieved to remove the low frequency interference;(2) Forhigh-frequency noises, we select the appropriate threshold function to use the waveletthreshold denoising method to remove high frequency interference;(3) Because of theactual pulse is often a compound signal which mix the power-line interference,baseline wanders and the muscle electrical noise, we proposed a mixed algorithmbased on the EEMD and wavelet threshold denoising. The experimental results shownthat compared with other methods, the denoising effect of this algorithm is obviouslyimproved.2. Aim at the problem of l feature extraction in the pulse signal, at the same timethis is the premise of realizing automatic analysis in pulse signal. For the main waveand beat wave recognition problems, a method of feature points detection based onstationary wavelet transform (SWT) and modulus maxima is proposed which realizedthe accurate positioning and extraction time-domain values; To dig deeper intotime-and-frequency domain features of the pulse, we extract the energy moment ofIMFs, wavelet energy coefficients matrix and energy characteristics based HHTmarginal spectrum. Thus realizes the extraction of multi-domain feature parameters inthe pulse signal accurately.3. By the support vector machine (SVM) classification method, achieving the classification of the two kinds of pulse signal. And according to principal componentanalysis (PCA) and independent component analysis (ICA), the influence of these twodimension reduction technique on the performance of the SVM classifier has carriedon the detailed study. Then characteristics classification algorithm based on the pulsesignal features and ICA-SVM is proposed. Through MP monitoring system,30of thenormal group and20of the hypertension of pregnancy group in actual pulse databaseare tested. Experimental results verified the the extracted feature valuse has stabilityand effectiveness, which can be used as the characteristics of classification, and alsoverified that the pulse signal classification algorithm proposed in this paper has afeasibility.