基于改进K-means聚类和霍夫变换的稀疏源混合矩阵盲估计算法

Blind Recovery of Mixing Matrix with Sparse Sources Based on Improved K-means Clustering and Hough Transform

混合矩阵的估计是稀疏源盲分离的关键组成部分,其估计精度直接影响到源信号的估计精度.本文首先针对K-means聚类算法依赖初始值选取的问题,将微分进化算法思想引入到K-means聚类算法中,提出了一种改进的K-means聚类算法.利用该算法,对稀疏源混合信号数据进行聚类,保证了聚类结果的鲁棒性.然后利用霍夫变换,对每一类数据的聚类中心进行修正,从而估计出混合矩阵,提高了混合矩阵的估计精度.仿真实验表明,相比于经典的稀疏源混合矩阵盲估计算法,本文算法具有更强的鲁棒性和更高的估计精度.

Blind mixing matrix recovery is one of the most important steps in blind separation of sparse sources, which impacts significantly on the recovery accuracy of source signals. A novel improved K-means clustering algorithm is proposed based on differential evolution,to avoid the partial convergence problem of the K-means algorithm. The proposed algorithm is applied to allocate the sparse mixture data to several clusters, thus guaranteeing the robustness of the clustering. Then the cluster centers are amended through Hough transform to recover the mixing matrix. Experimental results show that the proposed mixing matrix recovery algorithm has advantages of high robustness and accuracy compared with conventional algorithms.