结合KPCA和多尺度纹理的IKONOS遥感影像决策树分类  

The Decision Tree Classification Based on KPCA and Multi-scale Texture Using IKONOS Remote Sensing Image

作　　者：谢丽军[1] 张友静[2,1] 张子衡[1] 陈李家[1] 

机构地区：[1]河海大学水文水资源学院,南京210098 [2]河海大学水文水资源及水利工程国家重点实验室,南京210098

出　　处：《遥感信息》2010年第3期88-93,共6页Remote Sensing Information

基　　金：国家自然科学基金重点项目(40830639)

摘　　要：城市地物类型多样,空间分布复杂,具有很强的非线性特征。核主成分分析(KPCA)通过将特征空间映射到高维核空间,可以表达图像像素间的高阶关系,因而可以提取图像的非线性特征,同时提供一组相互独立的主成分。本文在加入多尺度纹理特征的基础上,以应用地物分布的空间细节信息;且利用核主成分分析(KP-CA)方法对光谱和纹理量提取非线性特征信息,增大类别之间的可分性;并结合决策树分类方法对IKONOS遥感影像分类。实验结果表明:KPCA能很好提取地物之间的非线性特征,结合KPCA和多尺度纹理的决策树分类方法能有效地提取地物类型,提取精度为79.3%,KAPPA系数为0.763.The urban features are strong non-linear character,multiple in type,and complex in spatial distribution.The Kernel Principal Component Analysis（KPCA） can express this high-level relationship of pixels and provide an independent KCP,by changing the feature space to the high dimensional kernel space.In the paper,by adding multi-scale texture features for application of the details of spatial distribution of features,KPCA method is used to extract the non-linear characteristics of urban features,increasing the separability between the categories;and decision tree classification method is used to the high-resolution remote sensing image to extract the urban features.The experimental results show that： KPCA can extract the good non-linear characteristics between the features,and combined with KPCA and multi-scale texture,the decision tree classification method can effectively extract the type of features,extraction accuracy of 79.3%,KAPPA coefficient of 0.763.

关 键 词：核主成分分析(KPCA) 多尺度纹理 IKONOS 决策树 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]