基于CART决策树分类的沙漠化信息提取方法研究  

Desertification Information Extraction Method Research based on the CART Decision Tree Classification

作　　者：吕利利[1] 颉耀文[1,2] 黄晓君[1] 张秀霞[1] 李汝嫣 

机构地区：[1]兰州大学资源环境学院,甘肃兰州730000 [2]兰州大学西部环境教育部重点实验室,甘肃兰州730000

出　　处：《遥感技术与应用》2017年第3期499-506,共8页Remote Sensing Technology and Application

基　　金：国家自然科学基金项目(41471163);国家自然科学基金重点项目(41530752);国家基础科学人才培养基金科研训练及科研能力提高项目(J1210065);中央高校优秀研究生创新项目(lzujbky-2016-247)

摘　　要：沙漠化是我国北方土地退化的主要形式之一,也是国内外研究中的重要环境问题。以民勤县为例,讨论了CART(Classification and Regression Tree)决策树在沙漠化研究中的应用,并使用Landsat8OLI遥感影像为数据源,构建了一种可行的用于研究区的沙漠化信息提取规则,进行地表沙漠化信息提取。结果表明:所构建的决策树模型结构简单,沙漠化提取效果较好;在研究区域达到87.70%的分类精度,Kappa系数为0.848 4,分类精度也较高。同时,归一化裸露指数(NDBI)和地表反照率(Albedo)是两个明显的沙漠化特征量,在沙漠化提取中起着重要作用。然而,CART决策树作为一种基于监督的分类方法,模型构建时,选择相对较高质量的训练样本和准确合理的输入端变量,可大大提高沙漠化信息的提取精度。Desertification is a main form of land degradation in northern China,and also an major environmental problem in China and foreign countries in recent years.Discussed the application of CART（Classification and Regression Tree）decision tree in desertification research in MinQin county,used the Landsat8 OLI remote sensing image data,and constructed a feasible rule to extract the desertification information.The result showed that the decision tree constructed not a simple and easily understand model,but a higher classification accuracy up to 87.7% and a Kappa coefficient is 0.848 4 in desertification information extraction.At the same time,the Normalized Difference Bare Index（NDBI）and the surface Albedo（Albedo）were two obvious variables in desertification,which played an important role in desertification extraction.However,the CART decision tree as a kind of classification method based on supervised,a relatively high quality training samples and a reasonable input variables is important for improving the information acquire accuracy in model construction of desertification information extraction.

关 键 词：遥感影像 CART决策树 沙漠化 信息提取 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置] P941.73[自动化与计算机技术—控制科学与工程]