基于集对分析的遥感图像K-均值聚类算法  

SPA-based K-means Clustering Algorithm for Remote Sensing Image

作　　者：谢相建[1] 赵俊三[1] 陈学辉[1] 袁思[1] 

机构地区：[1]昆明理工大学国土资源工程学院,昆明650093

出　　处：《国土资源遥感》2012年第4期82-87,共6页Remote Sensing for Land & Resources

基　　金：国家自然科学基金"面向对象的土地利用空间多尺度耦合机理研究"(编号:41161062)资助

摘　　要：基于欧式距离的K-均值聚类算法是一种硬分类(把每个待辨识的对象严格地划分到某个类中)方法,面对具有不确定性和混合像元特征的遥感图像数据,传统K-均值聚类算法很难得到满意的分类结果。为解决这一难题,将集对分析(set pair analysis,SPA)理论推广到遥感图像聚类算法,通过引入一个能统一描述同一性、差异性和对立性的同异反(identical discrepancy contrary,IDC)联系度,提出了基于IDC联系度的改进的K-均值聚类算法。该方法克服了传统K-均值算法硬分类的缺陷,可以有效地提高遥感图像聚类精度。对Landsat5 TM卫星数据的聚类分析实验表明,在含有混合像元的遥感图像地物覆盖分类中,改进的K-均值聚类方法的分类效果要优于传统K-均值聚类方法。K-means clustering algorithm is a kind of hard classification based on the Euclidean distance,with each data point assigned to a single cluster.Due to the uncertainty and mixed pixels in remote sensing image,it is difficult for the traditional K-means clustering algorithm to obtain satisfactory classification results.To overcome this drawback,the authors applied the SPA（set pair analysis）theory to the clustering algorithm of remote sensing image.The IDC（identical discrepancy contrary）connection degree model,which can descript unitarily the identity,discrepancy and opposition,was employed to improve K-means clustering algorithm.The improved algorithm has overcome the limitation of K-means clustering algorithm to certain extent.Clustering analysis experiments of Landsat TM image show that the improved K-means clustering algorithm is superior to K-means in classification accuracy of ground cover class components of mixed pixels.

关 键 词：集对分析 K-均值聚类算法 同异反联系度 遥感图像 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]