基于单极化TerraSAR-X影像提取建筑区研究  

Built-up Areas Extraction Based on Single Polarization TerraSAR-X Images

作　　者：蒋丹丹[1] 卢刚[2] 陈成[2] 

机构地区：[1]中国矿业大学环境与测绘学院,江苏徐州221116 [2]江苏省测绘工程院,江苏南京210013

出　　处：《地理与地理信息科学》2016年第1期60-65,F0003,共7页Geography and Geo-Information Science

基　　金：国家高技术研究发展计划(863计划)项目(2013AA122003)

摘　　要：基于多时相单极化TerraSAR-X影像,提取后向散射系数特征影像、变差函数纹理特征影像、干涉特征影像,并对其波段组合,在基于eCognition软件多尺度分割的基础上,选择训练样本对象,充分挖掘样本对象的光谱、纹理特征信息,通过建立特征分离性指标,降低特征维数,最后应用面向对象的分类方法实现了建筑区的自动提取,提取精度达93.50%。这一研究表明特征组合影像具有较大的分割优势,特征的优化选择在减少特征冗余的同时维持了较高的信息提取精度,且弥补了eCognition软件纹理特征计算慢的不足。This paper shows a object-oriented method that the extraction of urban built-up areas based on multi-temporal single polarization TerraSAR-X images.Firstly,the backscattering coefficient images,variogram texture image and interferometric coherent image were combined,and the combined image was segmented with a multiresolution segmentation algorithm based on eCognition software.And then choosing the train samples,extracting the spectral and textural features of sample objects,sample objects separability index was established in order to reduce feature dimensions.Finally,the built-up areas were extracted automatically with a object-oriented classification approach,and total accuracy is up to 93.50%.The results indicates that combined image has great advantage of segmentation,optimal selection of features reduces redundant features,maintains the high accuracy of information extraction,and makes up for the inadequacy that texture characteristics calculation is slow based on eCognition software.

关 键 词：TerraSAR-X影像 建筑区提取 影像分割 降低特征维数 面向对象分类 

分 类 号：P237[天文地球—摄影测量与遥感]