计算机工程 ›› 2019, Vol. 45 ›› Issue (12): 314-320.doi: 10.19678/j.issn.1000-3428.0053216

• 开发研究与工程应用 • 上一篇    

带标签约束的心肺音分离方法

朱俊霖, 王海平, 杨祖元   

  1. 广东工业大学 自动化学院, 广州 510006
  • 收稿日期:2018-11-23 修回日期:2019-01-02 发布日期:2019-12-10
  • 作者简介:朱俊霖(1996-),男,硕士研究生,主研方向为非负矩阵分解、心肺音分离;王海平,硕士研究生;杨祖元,教授、博士。
  • 基金项目:
    国家自然科学基金(61722304)。

Cardiopulmonary Sounds Separation Method with Label Constraint

ZHU Junlin, WANG Haiping, YANG Zuyuan   

  1. School of Automation, Guangdong University of Technology, Guangzhou 510006, China
  • Received:2018-11-23 Revised:2019-01-02 Published:2019-12-10

摘要: 心音信号和肺音信号在时频域上的混叠会影响听诊效果,而传统基于非负矩阵分解(NMF)的心肺音分离方法在获取参考信号过程中没有利用心音和肺音的标签信息,使得分离精度受限。为此,在NMF的基础上引入标签约束,提出一种心肺音分离方法。将心肺音特有的频率特性以标签形式加入到心肺音分离算法中,经NMF分解得到心音和肺音的参考信号。在此基础上,通过分析参考信号和原始信号的相关性完成聚类,采用时频掩码实现心肺音信号分离。实验结果表明,与传统NMF方法和带通滤波法相比,该方法获得的信噪比和相关系数较高,且能对真实心肺音混合信号进行有效分离。

关键词: 心肺音分离, 盲源分离, 非负矩阵分解, 聚类, 标签约束

Abstract: Due to the overlapping of cardiac and respiratory sound signals in time-frequency domain,the effect of auscultation can be affected.Besides,the traditional cardiopulmonary sounds separation method based on Non-negative Matrix Factorization(NMF) fails to utilize the label information of cardiopulmonary sounds in the reference signal acquiring process,causing limitation in separation accuracy.Therefore,by introducing a label constraint on the basis of NMF,we propose a cardiopulmonary sounds separation method.Firstly,the unique frequency characteristics of cardiopulmonary sounds are added to the cardiopulmonary sounds separation algorithm in label form.Then,the reference signals of cardiopulmonary sounds are obtained by NMF decomposition.Finally,the clustering is completed by analyzing correlation between the reference signals and the original signals,and the separation of cardiopulmonary sounds signals is also achieved through time-frequency mask.Simulation results show that compared with the traditional NMF method and bandpass filtering method,the proposed method has higher signal-to-noise ratio and correlation coefficient.Meanwhile,it can effectively separate the real cardiopulmonary sounds mixed signals.

Key words: cardiopulmonary sounds separation, blind source separation, Non-negative Matrix Factorization(NMF), clustering, label constraint

中图分类号: