期刊名称：ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
印刷版ISSN：2194-9042
电子版ISSN：2194-9050
出版年度：2009
卷号：XXXVIII-7/C4
页码：281-285
出版社：Copernicus Publications
摘要：Texture features in high resolution TerraSAR-X data were used for classification in this paper. In single band and single polarized SAR image, texture holds useful information for interpreting objects in urban area. In this paper, the gray level co-occurrence matrix (GLCM) was computed to extract texture images. We used contrast, energy, correlation and mean measures combination based on GLCM to characterize texture images. Window size is an important parameter for mapping textures. Larger windows lead to more stable texture features but tend to blur the edges, while smaller windows lead to erroneous boundary delineation and misclassify the boundary itself as an incorrect class. To get the appropriate window size, nine window sizes 3×3, 5×5, 7×7…19×19 were tested on the filtered SAR image. The transformed divergence (TD) distance was computed for comparing separability between two classes from vegetation, roads, buildings and water body. According to the TD distance, the windows with size smaller than 11×11 would lead to unstable seperability and not be enough to fully separate four classes. And when the windows were set from 11×11 to 19 × 19, the separability was stable and better. So, we adopted the 11×11 window size considering both separability of classes and boundary delineation. Then, SVM classification techniques were used. We had a conclusion that, with texture features as accessorial data, the accuracy has a great more improvement, which proves an effective method to classify high spatial resolution SAR image
