遥感影像中辫状河道提取的CART决策树分类方法研究  

Research on braided channels extraction method of remote sensing image based on CART

作　　者：罗信 闫奇奇 宋思涵 杜泓达 兰志玺 龙颖波[1] Luo Xin;Yan Qiqi;Song Sihan;Du Hongda;Lan Zhixi;Long Yingbo(School of Earth Sciences,Yangtze University,Wuhan,Hubei 430100,China)

机构地区：[1]长江大学地球科学学院,湖北武汉430100

出　　处：《计算机时代》2022年第8期6-9,共4页Computer Era

基　　金：长江大学大学生创新创业训练计划项目(Yz2020010,Yz2020032)。

摘　　要：为精确提取形态复杂的辫状河道水体范围,采用多信息合成的CART决策树分类法对俄罗斯勒拿河入海口的三角洲进行辫状河道自动提取实验。对预处理后的Landsat 8遥感影像进行NDWI和ISODATA分类操作,将这两种操作的结果影像与原始影像进行组合,构建特征数据集;利用CART分类工具对研究区进行分类,将分类后的水体范围输出为矢量图形。结果表明该方法的河道提取精度优于最大似然分类法,较好地实现遥感影像中辫状河道信息的提取。In order to accurately extract the water areas of braided river with complex shape,we use the classification and regression tree(CART)method with multi-information synthesis to carry out the automatic extraction experiment of braided river in the delta of the Lena River estuary in Russia.The preprocessed Landsat 8 remote sensing image is classified by NDWI and ISODATA,and the result images of the two methods are combined with the original images to construct the synthetized feature data set.Then the CART classification tool is used to classify the study area,and finally the classified water areas are output as vector graphics.The results show that the channel extraction accuracy of this method is better than the maximum likelihood classification(MLC)method,and can better extract the braided channel information in remote sensing images.

关 键 词：河道提取 CART决策树分类 多信息合成 辫状河道 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]