Research On Methods Of Object-oriented Built-up Areas Extraction Based On Single Polarization TerraSAR-X Images

Identification of built-up areas from satellite imagery can provide a crucial information layer in management for monitoring urban sprawl, urban planning and disaster assessment. Spaceborne radar imagery is at a large enough advantage in regions where environmental conditions impede the acquisition of optical image data. It has become the research focus that automated methodologies for extraction of builtup areas must be developed or adapted to account for the specific characteristics of synthetic aperture radar(SAR) data. In this paper, taking high-resolution(HR) singlepolarized TerraSAR-X(TSX) as the data source, methods for built-up areas extraction were analyzed from two points of view of object-oriented texture analysis and classification based on multiple features, the main content includes:(1) Strong backscattering behavior is one of the distinct characteristics of builtup areas in HR SAR image. However, in practical application, not all of built-up areas pixels are highlighted in the SAR image. Texture feature is an important characteristic differing from other land-cover types, taking three typical city scenes for experimental areas, this study provides a comparative analysis of different texture information extraction methods based on Matlab software platform. Specific texture features were extracted based on different methods, with the support of eCognition software,then rule sets were established to obtain binary image based on object-oriented Otsu algorithm with the support of eCognition software, realizing the automatic determination of the texture feature threshold and achieving the goal of rapid extraction of the outline of built-up areas.(2) In view of the problem of built-up areas extraction by using the texture feature based on single polarization TSX image, Considering built-up areas are in a relatively stable state in a certain period of time, Combining feature images based on the analysis of backscatter, interferometric coherence, speckle divergence and texture feature derived from multi-temporal single polarization TSX images, then the combined image was firstly segmented with a multi-resolution segmentation algorithm, selecting the training samples and fully exploit the attribute information of sample objects, then separability index was established considering feature distance within and outside classes in order to features optimization. Finally, built-up areas were extracted by using object-oriented multiple features classification method. The results indicate that the method proposed reduces redundancy and maintains the high accuracy of information extraction, and makes up for the inadequacy in a certain extent that texture feature calculation is slow based on eCognition software.(3) Design and implementation of experimental platform for features extraction and optimization were finished, Building up the knowledge rule sets for outlines extraction of built-up areas based on single polarization TSX images, developing the action library, customizing visual interfaces, setting up a reliable solution for outlines extraction of built-up areas. The solution can help workers to extract built-up areas contours quickly and easily.