基于聚类算法实现信号盲分类

本文解决了信号处理、工业控制等领域存在的非平稳信号盲分类问题。在聚类中广泛应用的K-Means算法及其它基于中心的聚类算法有两个共同的缺陷-需要预先确定类数目且随机初始化中心引起性能不稳定。本文提出的算法较好地解决了这两个问题，提高了算法稳定性，实现了非平稳信号盲分类。提取非平稳信号的小波系数作为聚类的样本空间，分析聚类结果的统计偏差以估计类的数目，采用调和均值准则进行分类。最后给出的仿真结果表明本文提出的方法较传统的K-Means算法明显降低分类错误率。

Blind Signals Classification Based on Clustering Algorithms

A method is proposed to deal with the problem of blind classification of non-stationary transients, which usually occur in the field of signal processing and industrial control. There are two well-known shortcomings within K-Means algorithm prevailingly used in data clustering and other center-based clustering algorithms-the number of clusters must be known beforehand and the dependence of the algorithm performance on the initialization of the initial centers. By means of the method proposed here, above two problems can be solved as to improve the stability and classify the non-stationary transients. The wavelet coefficients of the transients are extracted to form the clustering sample space. The number of clusters is estimated through analyzing the gap statistics of clustering results, and then using harmonic means criterion to perform further classifying. Simulation result shows that the proposed method can be used to greatly reduce the misclassification probability.