Identity recognition using heart sound based on HHT

A method of identity recognition using heart sound based on HHT and SVM was proposed. With the approach, the heart sound signals were decomposed using the empirical mode decomposition (EMD) to get stable IMF components at first, aiming at the non-stable and non-linear of heart sounds. Then, the IMFs containing the information of the first and second heart sounds were selected and the corresponding HHT instantaneous spectrum was drawn by Hilbert transformation. Subsequently, energy character vectors of spectrum were taken input support vector machine (SVM) as samples to establish the classifier. To improve the classification accuracy, SVM´s parameters c and g are optimized by particle swarm optimization (PSO). The experiment collects 120 heart sounds from 3 people to test the proposed algorithm, and the test results indicate that the HHT+PSO-SVM achieved upper recognition rate. The classification accuracy of the HHT+PSO-SVM algorithm reached 92.8%, and results demonstrate that the method has an encouraging recognition performance and identity recognition using heart sound is feasible.