面向对象方法支持下的土地利用分类  

Land Use Classification Supported by Object-based Methods

作　　者：郭鹏程 周志易[1] GUO Pengcheng;ZHOU Zhiyi(College of Civil Engineering,Hefei University of Technology,Hefei Anhui 230009,China)

机构地区：[1]合肥工业大学土木与水利工程学院,安徽合肥230009

出　　处：《北京测绘》2021年第5期616-621,共6页Beijing Surveying and Mapping

摘　　要：面向对象的影像分析技术在高分辨率影像地物信息的提取中有着重要应用。利用Sentinel-2高分辨率多光谱影像数据,以合肥市包河区作为研究区域,应用多尺度分割技术将影像分割成对象,并对特征空间进行选择和优化,基于面向对象分类方法提取出研究区域最近邻的六种典型地物,分类结果与面向像元的最大似然分类、支持向量机、神经网络的结果进行比较。结果表明:利用面向对象方法进行土地利用分类的总体精度88. 90%,Kappa系数为0. 857 9,优于三种传统的监督分类方法。证明了面向对象的影像分析技术在土地利用分类中的实用性。Object-based image analysis plays an important role in the extraction of high-resolution image features.In this paper,taking the Baohe District of Hefei City as the study area,the Sentinel-2 high-resolution multispectral image was used as the data source.The image was divided into objects by multi-scale segmentation technology,the feature space was selected and optimized.The object-based classification method based on nearest neighbor was used to extraction of six typical feature types in the study area.The classification results were compared with that of pixel-based maximum likelihood classification,Support Vector Machine(SVM),and neural network.The results showed that the overall accuracy of land-use classification using object-based method was 88.90%,and the Kappa coefficient was 0.8579,which was superior to the three traditional supervised classification methods and verified the practicability of object-based classification technology in land use classification.

关 键 词：Sentinel-2 多尺度分割 面向对象 土地利用分类 

分 类 号：P237[天文地球—摄影测量与遥感]