Multi-label classification of heart sound signals

Abstract

In recent years, with the development of heart sound classification technology, it has played an important role in the detection of congenital heart disease. However, in the traditional heart sound classification tasks, they are all two classification tasks. However, the heart sound signals are actually collected from five different locations during the collection process. Before we classify the normal and abnormal heart sounds, we should carry out the multi-classification task of the mixed heart sound signals in the collection area. In this paper, Mel cepstrum coefficient and power spectral density are taken as data sets for machine learning. We are committed to finding the best classification results, so we set two different labels, one is the multi-classification task of the location of heart sound signal acquisition area, and the other is the two-classification task of normal and abnormal heart sound signals. The accuracy rate.recall rate and F1 rate of heart sound signal recognition after subarea can reach 92.92%, 91.56% and 94.04%, which provides reference for clinical heart sound classification.