基于决策树的CBERS遥感影像分类及分析评价  

CBERS IMAGERY CLASSIFICATION BASED ON DECISION TREE AND DERFORMANCE ANALYSIS

作　　者：袁林山[1] 杜培军[1] 张华鹏[1] 张海荣[1] 

机构地区：[1]中国矿业大学地理信息与遥感科学系

出　　处：《国土资源遥感》2008年第2期92-98,共7页Remote Sensing for Land & Resources

基　　金：江苏省自然科学基金创新人才青年学术带头人基金项目(BK2006505);江苏省高等学校“青蓝工程”中青年学术带头人培养计划资助;中国矿业大学科技基金项目(2005B018)共同资助

摘　　要：以江苏省徐州市为研究区,以城市土地利用遥感分类为目标,采用CBERS多光谱数据的近红外波段、全球环境监测植被指数(GEMI)、归一化植被指数(NDVI)及主成分分析得出的第一和第二主成分作为分类的特征数据,基于先验知识和统计分析构建层次分类决策树,进而发展和改进了决策树交互式构建算法,实现了城市土地利用遥感分类。通过与最大似然分类器(MLC)和支持向量机分类器(SVM)分类结果的比较分析,表明基于多种特征的决策树分类器能够有效应用于CBERS遥感数据分类,在研究区具有良好的推广性。In order to explore the application of China - Brazil Earth Resources Satellite （CBERS） remote sensing data to urban land cover/land use analysis, the authors developed the decision tree classifier, whose generation strategy is discussed in detail in this paper. With Xuzhou city as the study area, five features, i. e. , near - infrared band, Global Environment Monitoring Index （GEMI） , NDVI, and the first and second principal components, were extracted and used for urban land use classification. On the basis of experiments, the decision tree was designed based on prior knowledge and statistical analysis, and a new interactive decision tree generation strategy was developed to optimize threshold selection. A comparison of the classification results with results of Maximum Likelihood Classifier （MLC） and Support Vector Machine （SVM） classifier shows that the decision tree classifier that uses multiple features is effective in land use classification from CBERS imagery.

关 键 词：中巴地球资源卫星(CBERS) 决策树 支持向量机 分类 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]