Extracting built-up areas from TerraSAR-X data using object-oriented classification method

International Symposium on Earth Observation for One Belt and One Road

地球科学;政治学;社会学;经济学

Wang, Suyun^1,2 ; Sun, Z.C.^2,3

China University of Geoscience (Beijing), Beijing

100083, China^1

Key Laboratory of Digital Earth Science, Institute of Remote Sensing and Digital Earth, Chinese Academy of Sciences, Beijing

100094, China^2

Hainan Key Laboratory Earth Observation, Sanya, Hainan

572029, China^3

关键词: Classification results;    Confusion matrices;    High resolution imagery;    Intensity information;    Object oriented classification;    Region growing algorithm;    Statistical indicators;    Texture information;   

Others  :  https://iopscience.iop.org/article/10.1088/1755-1315/57/1/012036/pdf
DOI  :  10.1088/1755-1315/57/1/012036

来源: IOP

【 摘 要 】

Based on single-polarized TerraSAR-X, the approach generates homogeneous segments on an arbitrary number of scale levels by applying a region-growing algorithm which takes the intensity of backscatter and shape-related properties into account. The object-oriented procedure consists of three main steps: firstly, the analysis of the local speckle behavior in the SAR intensity data, leading to the generation of a texture image; secondly, a segmentation based on the intensity image; thirdly, the classification of each segment using the derived texture file and intensity information in order to identify and extract build-up areas. In our research, the distribution of BAs in Dongying City is derived from single-polarized TSX SM image (acquired on 17th June 2013) with average ground resolution of 3m using our proposed approach. By cross-validating the random selected validation points with geo-referenced field sites, Quick Bird high-resolution imagery, confusion matrices with statistical indicators are calculated and used for assessing the classification results. The results demonstrate that an overall accuracy 92.89 and a kappa coefficient of 0.85 could be achieved. We have shown that connect texture information with the analysis of the local speckle divergence, combining texture and intensity of construction extraction is feasible, efficient and rapid.