欢迎访问林草资源研究

林业资源管理 ›› 2015›› Issue (4): 73-78.doi: 10.13466/j.cnki.lyzygl.2015.04.013

• 科学研究 • 上一篇    下一篇

基于高分一号影像光谱指数识别火烧迹地的决策树方法

祖笑锋1, 覃先林1, 尹凌宇1, 陈小中2, 钟祥清2   

  1. 1.中国林业科学研究院资源信息研究所,北京 100091;
    2.四川省林业信息中心,成都 610081
  • 出版日期:2015-08-28 发布日期:2020-12-01
  • 通讯作者: 覃先林(1969-),男,四川南溪人,副研究员,博士,硕导,主要从事植被变化及林火预警监测技术研究。Email:noaags@caf.ac.cn;qxl9157@126.com
  • 作者简介:祖笑锋(1988-),男,黑龙江肇州人,在读硕士,主要从事光学遥感影像处理和森林火灾监测方法研究。Email:zuxiaofeng_lky@163.com
  • 基金资助:
    民用航天预研项目“基于多源空间数据的森林火灾综合监测技术与应用示范”;国防科工局重大专项项目(21-Y30B05-9001-13/15)

Decision Tree Method for Burned Area Identification Based on the Spectral Index of GF-1 WFV Image

ZU Xiaofeng1, QIN Xianlin1, YIN Lingyu1, CHEN Xiaozhong2, ZHONG Xiangqing2   

  1. 1.Research Institute of Forest Resource Information Technique,the Chinese Academy of Forestry,Beijing 100091,China;
    2.Forestry Information Center of Sichuan Province,Chengdu 610081,China
  • Online:2015-08-28 Published:2020-12-01

摘要: 森林火灾发生后,为及时、准确地掌握森林受灾情况,利用高分一号卫星(GF-1)16m宽幅影像各波段反射率信息,结合计算的归一化植被指数(NDVI)、过火区识别指数(BAI)、阴影植被指数(SVI)、归一化差异水体指数(NDWI)和全球环境监测指数(GEMI)等5种光谱指数,构建森林火烧迹地识别决策树模型(CART);在选取的研究区对该模型方法进行验证,并与最大似然监督分类法和非监督分类(ISODATA)方法所得到的结果精度进行了对比分析,结果表明采用基于CART模型的决策树方法对火烧迹地识别结果精度较最大似然法总体分类精度提高了4.38%,Kappa系数提高了0.102 4,制图精度提高了14.96%,用户精度提高了8.50%;而采用ISODATA方法识别的火烧迹地的总体精度和Kappa系数都较低,制图精度和用户精度都没有达到1%。

关键词: 高分一号卫星影像, 森林灾害, 火烧迹地, 植被指数, 决策树模型

Abstract: This paper describes the technique to be needed for rapidly and accurately identifying the burned area by forest fires,following the catastrophic fires by the vegetation index CART decision tree methods using the wide coverage image of GF-1(GF-1 WFV).They were compared between the maximum likelihood classification of supervised and unsupervised classification(ISODATA),within burned area indexes,to improve the accuracy of the burned area,shaded vegetation index,global environment monitoring index,improved shadows and bare commission or omission burned phenomenon.The results showed that the decision tree classification method based on CART algorithms for burned area identification has significantly improved the overall accuracy by 4.38% compared with the maximum likelihood method;Kappa coefficient increased by 0.1024.GF-1 satellite imagery for unsupervised classification(ISODATA)identifies the burned area poorly,the overall accuracy and Kappa coefficient are low,the map making accuracy and user accuracy have not reached 1%.

Key words: GF-1 satellite images, forest disaster, burned area, vegetation index, the decision tree model

中图分类号: