Multitemporal RADARSAT-2 Polarimetric SAR Data for Urban Land Cover Classification using Support Vector Machine

This study investigates the capability of the RADARSAT-2 Polarimetric SAR data for urban land-cover mapping using a novel classification approach. Six-dates of RADARSAT-2 polarimetric SAR data were acquired during June to September of 2008 in the rural-urban fringe of the Greater Toronto Area. The major landuse/land-cover classes were high-density built-up areas, low-density built-up areas, roads, forests, parks, golf courses, water and several types of agricultural crops. In the proposed approach, support vector machine (SVM) is combined with the rule-based method for the object-based classification of the multitemporal polarimetric SAR data. First, various SAR polarimetric parameters are derived from coherency matrix as well as Pauli, Freeman and H/A/a decompositions of the SAR data. Second, the multi-resolution segmentations of the selected SAR features are performed to generate meaningful image objects. Then the image objects containing the SAR polarimetric features are classified using the SVM classifier.

The SVM classification results are further refined using a rule-based approach. Rules are built to recognize specific classes defined by the shape features and the spatial relationships within the context. Effectiveness of different SAR ploarimtric parameters for urban land cover classification are compared. The primary result shows that the object-based classification using SVM and rule-based approach is promising for urban land cover mapping. In addition, finer urban structures such as roads could be extracted and certain changing areas such as construction sites could be detected.