COMBINATION OF TERRASAR-X AND OPTICAL IMAGERY FOR LU/LC MAPPING USING AN OBJECT-BASED APPROACH

In this study, the impact of the use of backscattering intensity and texture features obtained from TerraSAR-X images for LULC classification of agricultural and forest areas, and its combination with features extracted from Landsat 7 EMT+ optical imagery is analyzed. The performance of texture descriptors on radar images is evaluated. After data pre-processing and the definition of classes in the study area, every object is described by means of a set of features computed from the TerraSAR-X and optical imagery, using a plot-based approach. Cadastral cartographic limits are employed for objects definition. Next, objects are classified using decision trees combined with boosting techniques. The classification results are compared to the LULC contained in the testing database, and the errors evaluated in terms of the different groups of variables, the source of data used, and their performance for the variety of classes considered. The classification results bring some possibilities and limitations of combining features from optical and radar imagery, evidence the complementary information provided by both types of data to face these applications.