摘要
针对谱聚类存在构造相似度矩阵时对尺度参数敏感以及处理多重尺度数据集效果不理想的缺陷,提出一种基于密度调整的改进自适应谱聚类算法.该算法将样本点所处领域的密度引入谱聚类,利用密度差来调整样本点之间的相似度,使其更符合实际簇类中样本点间的内在关系,在一定程度上解决了多尺度聚类问题;同时,通过样本点的近邻距离自适应得到尺度参数,使算法对尺度参数相对不敏感.仿真实验验证了所提出算法的有效性和优越性.
As spectral clustering is sensitive to the scaling parameter while calculating the affinity matrix and the result of clustering multi-scale dataset is not ideal, an improved adaptive spectral clustering algorithm based on density adjustment is proposed. The algorithm introduces local density of data into spectral clustering, using the density difference to adjust the similarity between sample points, which makes it more consistent with the data points’ internal relations of the clusters’ actual structure. So that it solves the multi-scale clustering problem to some extent. At the same time, the algorithm is relatively insensitive to the scaling parameter by using the distances between data points and their neighbor points to get the scaling parameter adaptively. Simulation experiment shows the effectiveness and superiority of the proposed algorithm.
出处
《控制与决策》
EI
CSCD
北大核心
2014年第9期1683-1687,共5页
Control and Decision
基金
国家自然科学基金项目(61273187)
教育部博士点新教师类基金项目(20120162120022)
湖南省科技计划项目(2012CK4018)
关键词
谱聚类
密度调整
自适应
尺度参数
多重尺度数据集
spectral clustering
density adjustment
adaptive
scaling parameter sensitivity
multi-scale dataset