| 利用TM数据提取干旱区土地覆被信息的方法比较 |
| |
| 引用本文： | 沙占江,马海州,李玲琴,樊启顺,黄华兵,杨海镇,曹广超.利用TM数据提取干旱区土地覆被信息的方法比较[J].干旱区地理,2005,28(1):59-64. |
| |
| 作者姓名： | 沙占江 马海州 李玲琴 樊启顺 黄华兵 杨海镇 曹广超 |
| |
| 作者单位： | 中国科学院青海盐湖研究,西宁,810008;青海师范大学地理与资源环境科学系,西宁,810008;中国科学院青海盐湖研究,西宁,810008;青海师范大学地理与资源环境科学系,西宁,810008 |
| |
| 基金项目： | 中国科学院知识创新重要方向项目(KZCX2-SW-118)，国家重点科技攻关项目(97-924-02-02A) |
| |
| 摘 要： | 以柴达木盆地香日德绿洲作为研究实验区，对该区域ETM遥感数据经过空间分辨率融合、主成分分析等方法进行空间信息增强及专题信息增强处理，组合最佳视觉背景图像，分别在不同背景图像上选择训练样本，利用最大似然法监督分类方法(MLC)、多空间尺度分层聚类(SSHC)和基于知识的模糊聚类方法(KFC)等分类器，分别用各自训练样本初始化各类别信息特征值，形成类别特征值模式库，分别以此为基础对待分样本进行分类，对初分类的结果经过类别合并、碎斑滤除以及重新编码赋色等分类后处理，得到最终分类结果及分类精度评价结果。从所获数据可以得出如下结论：从总体精度和Kappa值可知，SSHC和．KFC分类方法所获结果精度较高，总体精度比MLC分类结果约高于3％，SSHC之结果精度略高于KFC之结果；SSHC、KFC和MLC三种分类方法对该区域地表覆被信息的提取分类中，SSHC分类方法对耕地、石砾地、河滩和荒漠分类结果较好，KFC分类方法对耕地、沙地、河滩和荒漠分类结果较好，MLC分类方法对耕地、河滩和荒漠分类结果较好，三种分类方法对耕地、河滩和荒漠等三种地类的分类精度较高，用户精度都在80％以上，而对沙地和石砾地的分类结果其用户精度大都低于80％。
|
| 关 键 词： | TM数据 土地覆被 LUCC 遥感 分类方法 干旱区 |
| 文章编号： | 1000-6060(2005)01-0059-06 |
| 修稿时间： | 2004年3月12日 |
Study on the Methods of Deriving the Information of Land Cover in the Arid Areas by Using TM Data |
| |
| SHA Zhan-jiang,MA Hai-zhou,LI Ling-qin,FAN Qi-shun,HUANG Hua-bing,YANG Hai-zhen,CAO Guang-chao.Study on the Methods of Deriving the Information of Land Cover in the Arid Areas by Using TM Data[J].Arid Land Geography,2005,28(1):59-64. |
| |
| Authors: | SHA Zhan-jiang MA Hai-zhou LI Ling-qin FAN Qi-shun HUANG Hua-bing YANG Hai-zhen CAO Guang-chao |
| |
| Institution: | SHA Zhan-jiang1,MA Hai-zhou1,LI Ling-qin2,FAN Qi-shun2,HUANG Hua-bing1 2 YANG Hai-zhen2 CAO Guang-chao1,2 |
| |
| Abstract: | By taking the Xiangride Oasis as the study area, in this paper, the spatial information and the thematicinformation of land use and land cover in the study area are enhanced by using the ETM remote sensing data afterinosculating the spatial resolving power and analyzing the principal components. The optimal color images arecomposed by using some processed band data, the training samples are selected from the background images. Byusing the maximum likelihood supervision classification (MLC), multi-spatial-scale hierarchical clusteringalgorithm (SSHC) and geographic-knowledge-based fuzzy clustering (KFC), the training samples areseparately initialized to obtain the eigenvalues of all kinds information so as to develop the eigenvalue modedatabase. All pixels were classified by using the mode base to get the initial classified results, the finally classifiedresults and classification precision are derived by post-processing including the clumping, riddling, eliminationand recoding for the initial classified results. The conclusions are as follows: the classified accuracy of SSHC andKFC is 3% higher than that of MLC, and the classified accuracy of SSHC is slightly higher than that of KFC. Inthree methods for deriving the information of land use and land cover, SSHC is more ideal to class farmlands,gravel lands, flood lands and deserts, KFC is more ideal to class farmlands, sand lands, flood lands and deserts,and MLC is more ideal to class farmlands, flood lands and deserts. The accuracy of all three methods forclassifying farmlands, flood lands and deserts is higher than 80%, but it is lower than 80% in classifying sandlands and gravel lands. |
| |
| Keywords: | TM data land use/cover LUCC remote sensing classifying method arid area |
| 本文献已被 CNKI 维普 万方数据 等数据库收录！ |
|
