摘要
为解决三维模型聚类中存在的聚类结果对数据输入顺序和维度敏感的问题,将基于自适应谐振理论的ART2网络引入到模型聚类中。以Rand指数、调整Rand指数和互信息指数3种聚类有效性评价指标为依据,通过实验分析了ART2网络中a,b,c,d,θ五个参数对聚类有效性的影响,并给出了一组较优的参数组合。在此基础上,定性地分析了警戒系数对聚类结果的影响,其中包括最大聚类数的确定和聚类结果对输入顺序的敏感度。聚类结果验证了ART2网络在模型聚类上的可行性和实用性。
Clustering results were always sensitive to dimensionality and input sequence of data in Three-Dimensional（3D） model clustering approaches.To solve this problem,Adaptive Resonance Theory（ART） based ART2 neural network was introduced to 3D model clustering.Adopting three indices which included Rand indice,adjustment Rand indice and mutual information indice as a reference point for judging clustering validity,the effect of ART2 network＇s five parameters：a,b,c,d and θ on clustering validation was evaluated through clustering experiments.Furthermore,a set of superior parameters was presented.On that basis,the effect of vigilance parameter on clustering result was analyzed qualitatively,which included the definite of maximum clustering number and clustering sensitivity to input sequence of data.The clustering experimental tests demonstrated the feasibility and effectiveness of ART2 network on model clustering.
出处
《计算机集成制造系统》
EI
CSCD
北大核心
2011年第9期1865-1872,共8页
Computer Integrated Manufacturing Systems
基金
国家自然科学基金重点资助项目(70931004)
国家863/CIMS主题资助项目(2009AA04Z122)~~
关键词
三维模型
聚类分析
ART2网络
聚类有效性评估
数据挖掘
three dimensional model
cluster analysis
adaptive resonance theory 2 network
clustering validity assessment
data mining
