基于密度峰值和模糊聚类的欠定混叠矩阵估计  

Underdetermined mixing matrix estimation algorithm using FCM based on density peak

作　　者：魏爽 俞守庚 杨璟安 WEI Shuang;YU Shou-geng;YANG Jing-an(The College of Information,Mechanical and Electrical Engineering,Shanghai Normal University,Shanghai 200234,China)

机构地区：[1]上海师范大学信息与机电工程学院,上海200234

出　　处：《计算机工程与设计》2023年第4期1050-1057,共8页Computer Engineering and Design

基　　金：国家自然科学基金项目(61401145,61701306);上海市自然科学基金项目(19ZR1437600)。

摘　　要：传统聚类算法进行混叠矩阵估计时存在的聚类中心个数不确定和初始聚类中心的随机选取导致陷入局部最优的问题,为此提出一种基于密度峰值的改进模糊聚类算法进行欠定盲源分离的混叠矩阵估计。通过短时傅里叶变换提取信号在频域中的稀疏特性,利用寻找密度峰值聚类算法(clustering by fast search and find of density peaks,CFSFDP)自动获取聚类簇的数目和初始聚类中心;将获得的聚类数目和聚类结果作为模糊聚类算法(fuzzy c-means clustering,FCM)的初始输入参数,提高FCM聚类结果的精度。实验结果表明,该算法可以准确估计源信号的数目,相比传统FCM、层次聚类、基于密度峰值改进的粒子群等聚类算法,可以有效提高欠定盲源分离的混叠矩阵估计精度。To solve the problems of unknown cluster number and local optimum because of random initial centers of traditional clustering algorithms in the estimation of the mixing matrix,an improved fuzzy c-means clustering method based on density peak for underdetermined mixing matrix estimation was proposed.Short time Fourier transformation(STFT)was employed to obtain the sparsity characteristic of the observed signal.The CFSFDP algorithm was adopted to automatically obtain the number of clusters using the decision graph.The result of the CFSFDP algorithm was used as the initial clustering centers of the FCM.Simulation results show that the proposed method can effectively estimate the number of the sources.Compared with the traditional FCM algorithm,hierarchical clustering algorithm and particle swarm optimization algorithm based on density peak,the proposed method can improve the estimation accuracy of the mixing matrix for underdetermined blind source separation.

关 键 词：欠定盲源分离 混叠矩阵估计 稀疏表示 两步法 模糊聚类 密度峰值 语音信号盲分离 

分 类 号：TN911.7[电子电信—通信与信息系统]