基于PCA和多尺度纹理特征提取的高分辨率遥感影像分类  

Extraction of High Spatial Resolution Remote Sensing Image Classification based on PCA and Multi-scale Texture Feature

作　　者：刘友山[1,2] 吕成文[1,2] 祝凤霞[1,2] 高超[1,2] 

机构地区：[1]安徽师范大学国土资源与旅游学院,安徽芜湖241003 [2]资源环境与地理信息工程安徽省工程技术研究中心,安徽芜湖241003

出　　处：《遥感技术与应用》2012年第5期706-711,共6页Remote Sensing Technology and Application

基　　金：安徽省教育厅自然科学重点项目(KJ2010A154)

摘　　要：城市地物类型多样,空间分布复杂,而且地物具有多尺度性,不同的地物类型具有不同的纹理表达尺度。利用主成分分析法(PCA)对高分辨率遥感影像进行处理,以减少数据量、抑制噪声、突出主要信息。在此基础上,利用灰度共生矩阵法对PCA的第一主成分进行纹理特征提取,选择最佳的多尺度纹理组合进行决策树分类。实验结果表明:基于PCA和多尺度纹理特征的决策树分类方法能够有效地提取地物信息,分类精度达到82.4%,Kappa系数为0.78。The types of urban ground objects and their spatial distribution are complex. A jects are multi-scale,different types of urban ground objects have different texture scal nd the ground ob- e. The paper uses Principal Component Analysis（PCA） to deal with high-resolution remote sensing images in order to the quantity of data, suppress the noise, and highlight important information. On this basis, this pa tracts the texture features from the first principal component of PCA on basis of Gray Level Co-occ Matrix,and chooses the best combination of multi-scale textures to decision tree classification. The show that the method of the decision tree classification based on PCA and multi-scale texture can the types of ground objects effectively. The precision of classification is 82.4% and Kappa coeffi 0.78. reduce per ex- urrenee results extract cient is

关 键 词：主成分分析 多尺度纹理特征 高分辨率 

分 类 号：TP75[自动化与计算机技术—检测技术与自动化装置]