徐雯亮 王志刚

摘 要：基于面向对象的采样点和多种特征分类，本文以黄土高原典型区延安市吴起县为研究区，采用CART决策树分类、Bayes分类、随机森林分类方法提取土地利用类型，同时对比传统最大似然法分类，对分类结果进行分析评价。结果表明，面向对象方法得到的分类结果地类边界与实际相符，其中随机森林分类效果最优。对于不同地类，面向对象Bayes分类方法适用于采矿用地;面向对象随机森林分类对建设用地分类精度较高;面向对象方法及传统最大似然法对草地、林地分类精度均较好;基于面向对象的方法，采用隶属度函数对水域的分类较为准确，但对于细小水体的提取，传统最大似然提取效果更优。研究成果为中低分辨率遥感影像的黄土高原地区典型地物类型提取提供了方法参考。

关键词：黄土高原;面向对象;土地利用;遥感指数

中图分类号：F323.211文献标识码：A文章编号：1003-5168（2020）26-0155-04

Abstract： Based on object-oriented sampling points and multiple feature classifications， this paper took Wuqi County， Yan'an City， a typical area of the Loess Plateau， as the research area， and used CART decision tree classification， Bayes classification， and random forest classification methods to extract land use types， and compared with the traditional maximum likelihood classification， analyzed and evaluated the classification results at the same time. The results show that the classification result obtained by the object-oriented method is consistent with the actual situation， and the random forest classification effect is the best. For different land types， the object-oriented Bayes classification method is suitable for mining land; the object-oriented random forest classification has a higher classification accuracy for construction land; the object-oriented method and the traditional maximum likelihood method have better classification accuracy for grassland and woodland; based on the object-oriented method， the classification of waters by the membership function is more accurate， but for the extraction of small water bodies， the traditional maximum likelihood extraction is more effective. The research results provide a method reference for the extraction of typical feature types in the Loess Plateau from low- and medium-resolution remote sensing images.

Keywords： Loess Plateau;object oriented;landuse;remote sensing index

典型地区土地利用/覆被信息获取对分析环境变化及合理布局资源具有重要意义[1]。随着卫星传感器技术及分类算法的发展，面向对象分類最大限度地利用分割后对象的光谱、纹理、拓扑关系等特征，避免了传统分类单纯基于地物光谱信息造成的地物破碎、不连续的现象[2]。许多学者[3-6]对黄土高原地区的研究集中在高分辨率及特征地物的提取，对于黄土高原地区中低分辨率面向对象方法提取典型地物的研究还不多见。

本文基于采样点和多种特……