摘要
针对分类目的准确标识出有样本分布的空间区域位置,没有类分布先验知识,类数不能预先确定的情况,提出一种聚类新方法。该算法的初始类心为所有样本点,竞争获胜规则由最近邻改为阈值,竞争过程中同时进行类心合并。在样本数量较大时,提出网格中心法和网格采样法降低计算复杂度。实验结果证实该算法对初始设置和参数不敏感,且结束条件容易确定,在一定程度上聚类效果优于其它算法。
A new cluster method is presented, which is aimed at accurately identifying the location of the area with example points, without transcendental knowledge of class distribution and the number of class. In this method, initial centers of class are all the example points, and the competitive win rule is a threshold instead of the nearest neighborhood, and the combination of class centers is carded during the competing process. The method of grid center and method of grid sample are presented to lower the calculation complexity when the number of points is huge. The experimental result proves it＇s insensitive to initial conditions and parameters, and the end conditions are easy to be defined. It＇s also proves the performance of this algorithm is superior to others in a certain extent.
出处
《计算机工程与设计》
CSCD
北大核心
2006年第9期1656-1659,共4页
Computer Engineering and Design
关键词
合并
竞争学习
样本空间
位置标识
网格中心法
网格采样法
combination
competitive learning
example points space
location identification
method of grid center
method of grid sample
