img

官方微信

img

群号:冰川冻土交流群

QQ群:218834310

高级检索

冰川冻土 ›› 2018, Vol. 40 ›› Issue (4): 828-836.doi: 10.7522/j.issn.1000-0240.2018.0089

• 寒旱区水文水资源 • 上一篇    下一篇

基于随机森林模型的城市不透水面提取研究——以呼和浩特市为例

郜燕芳1, 李俊明2, 刘东伟1, 任周鹏3, 王楠楠4   

  1. 1. 内蒙古大学 生态与环境学院, 内蒙古 呼和浩特 010021;
    2. 山西财经大学 统计学院, 山西 太原 030006;
    3. 中国科学院 地理科学与资源研究所, 北京 100101;
    4. 河南大学 环境与规划学院, 河南 开封 475004
  • 收稿日期:2018-03-01 修回日期:2018-07-25 出版日期:2018-08-25 发布日期:2018-10-08
  • 通讯作者: 刘东伟,E-mail:liudw@imu.edu.cn. E-mail:liudw@imu.edu.cn
  • 作者简介:郜燕芳(1990-),女,内蒙古乌兰察布人,技术员,2017年在内蒙古大学获硕士学位,从事环境生态学研究.E-mail:gaoyf@lreis.ac.cn
  • 基金资助:
    国家自然科学基金项目(41571090;41201539;31560146)资助

Research on extraction of urban impervious surface based on random forest model: a case study in Hohhot

GAO Yanfang1, LI Junming2, LIU Dongwei1, REN Zhoupeng3, WANG Nannan4   

  1. 1. School of Ecology and Environment, Inner Mongolia University, Hohhot 010021, China;
    2. School of Statistics, Shanxi University of Finance and Economics, Taiyuan 030006, China;
    3. Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing 100101, China;
    4. College of Environment and Planning, Henan University, Kaifeng 475001, China
  • Received:2018-03-01 Revised:2018-07-25 Online:2018-08-25 Published:2018-10-08

摘要: 城市不透水面信息对于城市生态环境动态演化过程研究具有重要意义。以Landsat 8遥感影像为数据源,以呼和浩特市为实证区域,进行了随机森林模型应用于城市不透水面的提取研究,并与目前应用广泛的支持向量机模型进行了对比分析。研究表明:在不同的抽样比例训练样本条件下,随机森林模型对于城市不透水面的提取精度均优于支持向量机的提取精度;对于随机森林模型和支持向量机模型,70%的训练样本比例均为最佳训练样本抽样比例。在该抽样比例下,随机森林模型提取城市不透水面的总体分类精度为93.29%,Kappa系数为0.9051,支持向量机模型的总体分类精度为91.26%,Kappa系数为0.8757;随机森林模型对于城市裸土的识别度较高,能更好地将城市裸土和不透水面进行区分,而支持向量机模型对于城市裸土、不透水面和绿地的区分能力均弱于随机森林模型。综合而言,随机森林模型对城市不透水面的提取精度优于支持向量机模型,随机森林模型可以有效应用于城市不透水面提取领域,进一步丰富了城市不透水面提取方法体系构成。

关键词: 不透水面, 随机森林, Landsat 8, 支持向量机

Abstract: Urban impervious surface information is essential for the research of dynamic evolution of urban ecological environment. In this paper, based on Landsat 8 remote sensing image, taking Hohhot city as a case, the urban impervious surface has been extracted by using random forest model and compared with the widely applicable support vector machine model. The research results showed that (1) the accuracy of the extraction of urban impervious surface by random forest model is better than that of the support vector machine model under the condition of various proportion of sampling training samples; (2) 70% of the proportion of training sample is the most optimal for either random forest model or support vector machine model. On the basis of the most optimal proportion of training sample, the overall classification accuracy of extraction of urban impervious surface with random forest model is 93.29% with a Kappa coefficient of 0.9051. The classification accuracy of the support vector machine model is 91.26% with a Kappa coefficient of 0.8757; (3) the random forest model has a more sensitive recognition to urban bare soil, and can better distinguish the urban bare soil from the impervious surface, while the support vector machine model cannot accurately distinguish the bare soil, impermeable surface and the green space. In short, the random forest model can more effectively and accurately extract urban impervious surface from Landsat 8 image and be applied in the field of extracting urban impervious surface, enriching the method system of extracting urban impervious surface.

Key words: impermeable surface, random forest, Landsat 8, support vector machine

中图分类号: 

  • TP751