基于ETM+图像和决策树的水体信息提取——以鄱阳湖周边区域为例  

Water Body Extraction Based on Decision Tree and ETM+ Imagery: A Case Study of Poyang Lake Region

作　　者：程晨[1] 韦玉春[1] 牛志春[2] 

机构地区：[1]南京师范大学地理科学学院,南京210046 [2]江苏省环境监测中心,南京210036

出　　处：《遥感信息》2012年第6期49-56,共8页Remote Sensing Information

基　　金：江苏省普通高校自然科学研究计划项目资助(09KJA420001);国家自然科学基金(40771152);江苏高校优势学科建设工程资助项目

摘　　要：基于2000年1月27日的Landsat ETM+图像,比较并分析了鄱阳湖周边区域约1421km2范围内9种典型水体提取算法的提取结果。针对单一提取算法容易产生混淆地物的问题,根据图像的缨帽变换和新的波段组合DW=EWI×(b2+b3)/(b4+b5)构建了决策树水体提取模型。其中,使用缨帽变换中的亮度分量区分冰雪和沙地,绿度分量去除林地,湿度分量增强水体信息,使用波段组合DW进一步增强水体和其他地物的差异。2000年的ETM+图像的水体提取和2003年2月20日ETM+图像的模型验证结果表明:本文的决策树模型较好将水体与河滩、冰雪、沙地、阴影和林地等地表覆盖分离开来,弥补了当前单一水体提取指数和监督分类的不足,提取的水体更为完整、准确。In this paper,a region about 1421km2 around Poyang Lake was chosen as the study area and one Landsat ETM＋ image acquired on January 27,2000 was used to build decision tree model to extract the water body information.Nine results of water body extraction were firstly compared according to the previous typical extraction algorithms.Then,by considering the deficiency in these existing algorithms,a decision tree model combing the K-T transformation（K-T） and a new band DW=EWI×（b2＋b3）/（b4＋b5） was built in this research.In this model,Brightness and Greenness of K-T were used to distinguish ice,sand,woodland from water body respectively,Wetness of K-T was used to further enhance the water information in image,and band combination DW was proposed to increase the differences between water body and other land covers.The decision tree model was used to extract water body in ETM＋ image acquired in 2000 and the extracted result was further verified by another ETM＋ image acquired on February 20,2003.The result shows that the model can effectively separate water body from flood land,ice,sand,woodland and shadows,and obtain a higher extraction precision than other traditional water extraction algorithms and the supervised classification method.

关 键 词：水体提取 ETM+图像 决策树 遥感信息提取 鄱阳湖 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]