Science Service System

Summary of Proposal MTH2531

TitleImproving Algorithms of Change Detection by Utilizing Staring Spotlight TerraSAR-X Data
Investigator Thiele, Antje - Karlsruher Institut für Technologie (KIT), Institut für Photogrammetrie und Fernerkundung (IPF)
Team Members
Dipl.-Ing. Dubois, Clémence - Karlsruher Institut für Technologie (KIT), Institut für Photogrammetrie und Fernerkundung (IPF)
Dipl.-Ing. Boldt, Markus - Fraunhofer IOSB - Institut für Optronik, Systemtechnik und Bildauswertung, Abteilung Szenenanalyse
Prof. Dr. Hinz, Stefan - Karlsruher Institut für Technologie (KIT), Institut für Photogrammetrie und Fernerkundung (IPF)
Dr. Schulz, Karsten - Fraunhofer IOSB, Institut für Optronik, Systemtechnik und Bildauswertung
Kuny, Silvia - Fraunhofer IOSB, Scene Analysis
SummaryThe basic idea of our proposal is to test the new product Staring Spotlight Mode in three different studies of incoherent change detection. All three studies are briefly summarized., whereby the executive summaries of first and second study are identical to two ongoing TerraSAR-X proposals submitted by the PhD students. 1. Study The basic idea of the proposal is to detect changes in buildings shape and height after an event such as a natural disaster by use of radargrammetry. As such an event is not predictable; the chosen test site is located in the agglomeration area of Paris, France, where building demolitions are planned within the framework of an urban reconversion project. In this project, two different data sets will be used: interferometric data for the pre-event analysis, and radargrammetric data for the post-event analysis. For the post-event analysis and change detection approach, which is the subject of this proposal, high resolution TerraSAR-X data are required, in order to examine the potential of this acquisition mode for urban area. The change detection approach will start with the extraction of building features in both interferometric and radargrammetric dataset. By use of these features and of potentially underlying information about the original building footprint, a filtering will be performed in both data sets independently. Then, we will extract new features in the filtered data, which can be more relevant for change detection. Using these features, a change detection approach will be implemented in order to determinate the changes that occurred at building location. Finally, this approach will be validated through several quality assessments. 2. Study In the last few years, change detection based on remote sensing data has become a highly frequented field of research with multiple applications for practical use. To detect changes between temporarily different satellite images is of interest for example in terms of urban monitoring and disaster management. The planned approach allows the fully automatic detection of small-scaled changes in time series of SAR amplitude image data. The detection of other scaled changes will be investigated. Furthermore, the categorization of the detected changes will be an important part of the study. For this classification step, the time series analysis will play a significant role. Amongst others, seasonal effects can be used to get conclusions on the change category. Also the method for hierarchical image interpretation given by the morphological attribute profiles can be used to cluster similar changes between SAR image pairs. 3. Study If natural disasters such as earthquakes strike urban areas, a fast emergency response is most crucial. In order to get a good overview of the affected area, SAR sensors are the systems of choice. However, the extraction of suitable information from SAR data is not an easy task, especially if no pre-event data of the area are available. It has been noted before that areas of destroyed buildings show a higher backscattering intensity in SAR data than their surroundings. This is caused by heaps of debris from the destroyed building. This change in the backscattering intensity, but also distinguished differences in the texture of heaps due to small rubbles resulting in formations of many dihedral and trihedral corner reflectors, are of main interest. The concept of the study contains the matching of features extracted or learned from simulated SAR data with the signature of real measured data. In a first step, different damage models of buildings have to be modelled and studied. Second, based on the different modelled heaps of debris, the simulation work has to be done considering different parameter settings. Based on this simulated SAR data, feature extraction can be started. Additionally, the most reliable features have to be characterized and tested on real measured SAR data.

Back to list of proposals

© DLR 2004-2016