Investigation on the separability of slums by multi-aspect TerraSAR-X dual-co-polarized high resolution spotlight images based on the multi-scale evaluation of local distributions

Following recent advances in distinguishing settlements vs. non-settlement areas from latest SAR data, the question arises whether a further automatic intra-urban delineation and characterization of different structural types is possible. This paper studies the appearance of the structural type “slums” in high resolution SAR images. Geocoded Kennaugh elements are used as backscatter information and Schmittlet indices as descriptor of local texture. Three cities with a significant share of slums (Cape Town, Manila, Mumbai) are chosen as test sites. These are imaged by TerraSAR-X in the dual-co-polarized high resolution spotlight mode in any available aspect angle. Representative distributions are estimated and fused by a robust approach. Our observations identify a high similarity of slums throughout all three test sites. The derived similarity maps are validated with reference data sets from visual interpretation and ground truth. The final validation strategy is based on completeness and correctness versus other classes in relation to the similarity. High accuracies (up to 87%) in identifying morphologic slums are reached for Cape Town. For Manila (up to 60%) and Mumbai (up to 54%), the distinction is more difficult due to their complex structural configuration. Concluding, high resolution SAR data can be suitable to automatically trace potential locations of slums. Polarimetric information and the incidence angle seem to have a negligible impact on the results whereas the intensity patterns and the passing direction of the satellite are playing a key role. Hence, the combination of intensity images (brightness) acquired from ascending and descending orbits together with Schmittlet indices (spatial pattern) promises best results. The transfer from the automatically recognized physical similarity to the semantic interpretation remains challenging.