Identification Algorithm Based On Heart Sound Analysis

With the modern society’s rapid development of communication technology and network,human physical and virtal space are expanding constantly, and human being puts forward to ahigher request for human identity recognition’s accuracy, security and applicability. Thetraditional identification methods have been far from meeting this request, so we should seek formore safe, reliable and convenient ways to use for identifaication. Biometrics refers to the use ofcomputer technology to acquire biological characteristics for identification. Compared to thetraditional identification methods, these technologys which are based on biological characteristicsidentifications have the advantage of good performance for secutity, not been easily forged orstolen, not been easily forgotten or missing, and carried available at anytime and anywhere. Thesebiological characteristics identification technology have acquired preliminary promotion, but stillhave faced many challenges in practical application. Security is one of the prominent problems.Hence, exploring a more secure biometric method is becoming one of the outsanding issues in thearea of human identification.This paper mainly studies the identification algorithm which based on heart sounds’analysis.After analyzing the heart sounds’mechanism and its characteristics, we respectively haveproposed the identification methods based on the margical spectrum and the Vector quantization(VQ), the Mel Frequency Cepstrum Coefficients with its first order differential coeddicients andthe Gaussian Mixture model (GMM). In the experimental stage, we have built a heart soundsdatabase using the acquisted heart sounds samples, and those algorithms have been tested basedon this database, eventuall got a high recognition rate. The main works of this paper are asfollows:1. Firstly, expound the generation mechanism of heart sounds, introduce some commonbiometrics, and then review the history and current situation of heart sounds identificationtechnology at home and abroad.2. An identificational algorithm based on the signal margical spectrum and the VectorQuantization is proposed. Firstly, study the basic method of extracting the Margical spectrumwhich is used as characteristic parameters, and then describe the basic principle of the VectorQuantization. Secondly, propose the improved code split generation method based on commonlycode generation methods. Lastly, give the concrete procedures of recognition algorithm based onthe analysis of the specific heart sounds.3. An identificational algorithm based on the Mel Frequency Cepstrum Coefficients with its first order differential coeddicients and the Gaussian Mixture model is proposed. Firstly, study thefeature extraction of Mel Frequency Cepstrum Coefficients with its first order differentialcoefficients. Secondly, describe the basic principle of the Gaussian Mixture model and study thespecific estimation algorithm of model parameters. Lastly, give the concrete procedures ofrecognition algorithm based on the analysis of the heart sounds.4. In the experimental stage, we acquire heart sounds, build a heart sounds database, and testthe algorithm performance of margical spectrum with VQ and MFCC,ΔMFCC with GMM.According to the involved specific parameters, we spectively set different values for thoseparameters to improve the algorithms’performance, and achieve a high recognition rate.The research in this paper provides a more safety way of using biological characteristics,lays a theoretical basis and gives a technical support for the embedded heart sounds identificationsystem’s development in the next stage.