基于样本迁移的干旱区地表覆盖快速更新  

Sample Transferring based Fast Land Cover Updating in Arid Land

作　　者：盖一铭 阿里木·赛买提[1,2,3] 王伟 吉力力·阿不都外力 Gai Yiming;Samat Alim;Wang Wei;Abuduwaili Jilili(State Key Laboratory of Desert and Oasis Ecology,Xinjiang Institute of Ecology and Geography,Chinese Academy of Sciences,Urumqi 830011,China;University of Chinese Academy of Sciences,Beijing 100049,China;Chinese Academy of Sciences Research Center for Ecology and Environment of Central Asia,Urumqi 830011,China)

机构地区：[1]中国科学院新疆生态与地理研究所荒漠与绿洲生态国家重点实验室,新疆乌鲁木齐830011 [2]中国科学院大学,北京100049 [3]中国科学院中亚生态环境研究中心,新疆乌鲁木齐830011

出　　处：《遥感技术与应用》2022年第2期333-341,共9页Remote Sensing Technology and Application

基　　金：国家自然科学基金项目“样本与特征迁移的中亚典型城市覆被精细分类方法研究”(42071424);中国科学院战略性先导专项“咸海退缩产生的盐尘及其环境影响”(XDA2006030102);中国科学院青年创新促进会(2018476)。

摘　　要：阿姆河三角洲作为典型干旱区,干旱胁迫和次生的盐胁迫决定了本地区生态环境的复杂性和独特性,给遥感地表覆盖制图带来一定的困难。在土地利用/覆盖(LULC)遥感图像分类任务中,数量大、质量高、成本低的样本和速度快、性能稳定的分类器是高效实现高精度分类的关键。在一些偏远地区开展土地利用/地表覆盖遥感图像分类依然面临着标记样本空间上稀疏、时间上不连续甚至是缺失,人工收集成本高等问题。为此,结合最优树集成和样本迁移的思想,构建了一种高效的地表覆盖自动更新的新方法。该方法通过变化检测在历史产品上的同期影像上进行样本标签的标记,并将过去的地表覆盖类型标签转移到同源目标影像上,使用最优树集成(Ensemble of optimum trees,OTE)完成地表覆盖自动分类。根据阿姆河三角洲地区地表覆盖分类试验结果,表明该方法可以提取有效的地表覆盖标签,并能较高精度发实现土地利用/地表覆盖的自动分类更新。Amu river delta,as a typical arid land,was threatened by drought and salination,which contribute to the complexity and specificality of its ecological environment.In the Land Use/Land Cover(LULC)Remote Sensing(RS)image classification tasks,collecting large number of high quality samples at low-cost and a high efficient and robust classifier are always the crucial factors to obtain high-accuracy classification results.Howev⁃er,it was still problems facing RS imageries classification in some remote areas that marked samples were sparsely distributed,timely dissected or even intermittent,and manual tasks for field sampling cost high.In this end,a new frame of automatic land cover classification based on ensemble of optimum trees and sample transfer was promoted in this paper.In this frame,sample labels were marked on the historical image which is same time and source with the product,then these labels were transferred into targeted RS image.Then,OTE meth⁃od classification was performed.According to the results in this paper,the OTE with sample transferring based method can extract land cover labels efficiently and update LULC map in a fine accuracy.

关 键 词：样本迁移 最优树集成 变化检测 地表覆盖变化 遥感图像分类 干旱区地表覆盖 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]