基于潜狄利克雷分配的遥感影像聚类算法聚类个数敏感性分析
首发时间:2012-11-30
摘要:本文概要叙述了基于狄利克雷分配(Latent Dirichlet Allocation, 即LDA)的遥感影像聚类算法的原理和实现步骤,重点分析了遥感影像聚类算法中聚类个数的敏感性。首先,文章以影像处理常用软件ENVI中的Kmeans算法为基础,可视化对比了不同聚类数情况下各类地物的聚类效果,通过计算聚类结果平均似然值对聚类个数的敏感性进行了分析。然后,文章从聚类数值、函数目标、聚类效果熵值评价各个方面,综合对比分析了三种基于LDA的影像聚类方法的聚类个数敏感性,通过分析的结论来探讨优化聚类个数的方法。
For information in English, please click here
The Sensibility Analysis of the Number of Clusters in Image Clustering Methods based on Latent Dirichlet Allocation
Abstract:In this paper, the theory and steps of satellite image clustering methods based on LDA is briefly introduced. The sensitivity of the number of clusters in the methods is mainly analyzed. Firstly, the clustering effect over various geo-objects with different number of clusters is analyzed based on the Kmeans in ENVI. Then, comprehensive evaluation is done over three kinds of image clustering methods based on LDA by using the Entropy evaluation. Finally, the methods and problems to determine the optimal number of clusters are discussed.
Keywords: satellite image clustering LDA the number of clusters sensitivity analysis
论文图表:
引用
No.****
同行评议
共计0人参与
勘误表
基于潜狄利克雷分配的遥感影像聚类算法聚类个数敏感性分析
评论
全部评论0/1000