面向对象, 多尺度, 光谱差异, CART决策树, 沙地提取," /> 面向对象, 多尺度, 光谱差异, CART决策树, 沙地提取,"/> object-oriented, multi-scale, spectral differences, CART decision tree, sand extraction,"/> 基于CART决策树的沙地信息提取方法研究
收藏 ｜ 设为首页 广告服务 ｜ 联系我们 ｜ 在线留言
高级检索

干旱区地理 ›› 2019, Vol. 42 ›› Issue (5): 1133-1140.doi: 10.12118/j.issn.1000-6060.2019.05.19

• 地球信息科学 • 上一篇    下一篇

基于CART决策树的沙地信息提取方法研究

张睎伟1，王磊2，汪西原1,3   

  1. 1 宁夏大学物理与电子电气工程学院，宁夏 银川 750021； 2 西北土地退化与生态恢复省部共建国家重点实验室培育基地，宁夏银川750021；3宁夏回族自治区沙漠信息智能感知重点实验室，宁夏 银川 750021
  • 收稿日期:2019-02-10 修回日期:2019-05-24 出版日期:2019-09-25 发布日期:2019-09-19
  • 通讯作者: 汪西原
  • 作者简介:张睎伟（1993-），男，硕士，研究方向为遥感图像处理研究. E-mail：zhang_xiwei@163.com
  • 基金资助:
    国家自然基金（41561087）

Sand information extraction method based on CART decision tree

ZHANG Xi-wei1,WANG Lei2,WANG Xi-yuan1,3   

  1. 1 School of Physics and ElectronicElectrical Engineering,Ningxia University,Yinchuan 750021,Ningxia,China;

    2 Breeding Base for State Key Laboratory of Land Degradation and Ecological Restoration in Northwest China, Yinchuan 750021,Ningxia,China;3Ningxia Hui Autonomous Region Desert Information Intellisense Key Laboratory,Ningxia University,Yinchuan 750021,Ningxia,China

  • Received:2019-02-10 Revised:2019-05-24 Online:2019-09-25 Published:2019-09-19

PDF (PC)

292

赞

摘要： 为研究沙地信息提取的方法，采用基于CART决策树的面向对象方法，提取中卫市沙坡头区的沙地信息。首先对研究区进行多尺度分割和光谱差异分割得到对象层，然后选择合适的提取特征和训练样本点，最后输入选择的提取特征和样本点生成CART规则树，并对地物进行分类，提取出沙地信息。结果表明：采用面向对象的CART决策树方法提取沙地信息具有较高自动化程度和精确度，依此构建的CART决策树总体分类精度可达到77%，是最近邻分类结果的1.12倍，支持向量机分类结果的1.57倍，此外，NDBI（归一化裸露指数）、GSI（粒度指数）和SWIR 2（第七波段）均值可以成功的将沙地、戈壁和裸岩石砾地三个易混地物区分开来，是沙地提取过程中三个重要的特征指数。

关键词: font-size:10.5pt, 面向对象')">">面向对象, 多尺度, 光谱差异, CART决策树, 沙地提取

Abstract:  

This paper used the object-oriented method and CART decision tree method to extract the sand information with high degree of automation and comprehensive extraction features.The main research process is as follows: (1) Select the study area and preprocess the image of the study area. (2) Use multi-scale segmentation and spectral difference segmentation to obtain the object layer. (3) Select rich extraction features and training sample objects. (4) Training features and sample objects to get the CART rule tree. (5) Apply all objects to the rule tree to get the classification result. (6) Compare the Nearest neighbor and Support vector machine classification results.Finally,Compared with the current research on extracting sand information by CART decision tree.The overall classification accuracy reached 77%,which is 1.12 times of the Nearest neighbor classification result,1.57 times of the support vector machine classification result.In addition,normalized diffevence bare index (NDBI), granularity size index (GSI) and the seventh band (SWIR 2) can successfully distinguish three easily mixed objects of sand, Gobi and bare rock,which are three important characteristic indexes in the process of sand extraction.The experiment has proven this method is a feasible sand extraction method for actual desertification monitoring.

Key words: font-size:10.5pt, object">objectfont-size:10.5pt, -">-font-size:10.5pt, oriented')">">oriented, multifont-size:10.5pt, -">-font-size:10.5pt, scale')">">scale, spectral differences, CART decision tree, sand extraction