基于指数的遥感影像决策树分类方法  

Decision Tree classification of remote sensing images based on index

作　　者：嘎力巴[1,2] 臧淑英 李苗[1,2] 吴长山[1,2] 

机构地区：[1]哈尔滨师范大学地理科学学院,黑龙江150025 [2]黑龙江省普通高等学校地理环境遥感监测重点实验室,哈尔滨150025

出　　处：《环境与发展》2016年第5期43-48,共6页Environment & Development

基　　金：国家自然科学基金(41571199);黑龙江省自然科学基金项目(ZD201308);黑龙江省自然科学基金青年项目(QC2016050)

摘　　要：本研究以哈尔滨市为研究区,采用Landsat-8多光谱影像为数据源,计算出水体指数、植被指数、建筑指数和土壤指数等共20个地物光谱指数;基于指数的决策树分类方法提取土地覆被类型,并对分类结果进行精度验证,对比分析不同指数对分类精度的影响。经过精度对比、筛选五组数据作为变量用决策树方法进行分类,并与于单纯地物光谱的(普通)决策树分类结果和最大似然法相比较。结果显示:五组基于指数的决策树分类结果都比普通决策树分类和传统的最大似然法分类精度高,分类精度最高的一组与以上两种分类结果相比,总精度分别提高了2.59%和9.55%,Kappa系数分别提高了0.08和0.15。本研究呈现了基于指数的决策树分类方法在土地利用信息提取中的优势,为更好协调哈尔滨市土地利用与城市扩展提供研究依据。In order to study the advantage of decision tree based on index in land use information extracted, taking Harbin as a resample,there are 20 spectral indexes was calculated,including water index,vegetation,construction and soil type index and so on,using Landsat-8 mulispectral images. According to the validation sample verifies the accuracy of the result of the classification,and then comparison and analysis the influence of different index on classification accuracy.Based on this,the higher precision index was classified into 5 groups to extract the area land use type,the classification result compared with decision tree classification based on spectral and maximum likelihood method.The result indicates that five groups based on the index of decision tree classification precision were better than the decision tree classification based on spectral and the maximum likelihood classification,one group was highest among five groups, which total precision improved by 2.59% and 9.55%,the Kappa coefficient increased 0.08 and 0.15,respectively.This paper presents the advantage of the decision tree based on index in land use information extraction,which provide basic data for better coordination of Harbin city of land use and urban expansion.

关 键 词：决策树分类 指数 信息提取 土地利用 

分 类 号：X144[环境科学与工程—环境科学]