Heart sound recognition method of congenital heart disease based on improved cepstrum coefficient features

Abstract

The classification of heart sounds plays an important role in the detection of congenital heart disease. In recent years, the classification of heart sounds has made some progress, but it is mainly based on traditional acoustic features, which may be insufficient for heart sounds and easily influenced by complex and changeable environmental factors. In this paper, aiming at the traditional Mel cepstrum coefficient (MFCC), an improvement of heart sound signal characteristics is proposed, and a new window function expression is proposed in the windowing link of the extraction process. The data source of our 2016 Heart Sound Challenge serves as the data set. Finally, the new MFCC is used for feature learning and classification tasks, and compared with the traditional MFCC. A variety of recognition algorithms show that the average accuracy of the improved MFCC classification and recognition reaches 93.52%.