Kombinierte Objektdetektion und physikalische Modell-Inversion für PolINSAR-Bilddaten

• Applicant: Professor Dr.-Ing. Olaf Hellwich
• Subject Area: Geodesy, Photogrammetry, Remote Sensing, Geoinformatics, Cartography
• Term: from 2011 to 2015
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 190378420

• Final Report Year: 2015

Final Report Abstract

In this work we have studied the combination of object detection and physical parameter inversion from PolInSAR (Polarimetric Interferometric SAR) data. We have demonstrated how to use physical parameters as features for object detection, as opposed to purely image processing methods that rely solely on image based features. We also have demonstrated how object detection methods could help refine previously estimated physical parameters by exploiting contextual information. Furthermore, we have considered the combination of different types of physical (polarimetric and interferometric) containing complementary physical information. As a preliminary step, a literature review and an examination of the available data allowed us to identify the physical parameters to invert and the type of objects to detect. Given our data, we have chosen to invert the height of scatterers from surfaces and to detect geometrical structures defined by building parts and ground. This analysis has been possible by the application of Multi-Baseline Polarimetric Interferometric spectral methods, also well-known as SAR tomography. Then, we have developed a tomographic processing chain allowing to perform three operations: - Application of the MUSIC algorithm to invert scatterer heights from PolInSAR images. - Design of a new algorithm called TomoSNI in order to retain only meaningful points leading to outlier free point clouds. - Development of a new algorithm called TomoSeed that performs automatically the object extraction by condidering locally planar geometric primitives describing ground and buildings. In order to complement this detection, we have considered the extraction of polarimetric features to improve the separation between buildings and ground. To do so, we have developed a new polarimetric speckle filter called PolSAR-BLF allowing an improved estimation of the polarimetric covariance in terms of noise reduction and edge preservation. This method has been shown to outperform other state-of-the-art methods and has been proven useful to obtain noise free semi-supervised classification of objects in PolSAR images. We have also studied the limitations of PolSAR data for the characterization of buildings allowing us to define lines for future research. The PolSAR-BLF filter has been released to the public in the form of an open-source package, available at https://github.com/odhondt/PolSAR-BLF. With the rise of new sensors such as SENTINEL1, TerraSAR-X and FSAR and the increase in size of new datasets, automatic pattern recognition methods are an essential tool to reduce the recquired human interaction. The results obtained are encouraging and open a new way in the combination of computer vision methods with physical inversion techniques from multidimensional SAR data.

Publications

• “Automatic extraction of geometric structures for 3D reconstruction from tomographic SAR data”. In: Geoscience and Remote Sensing Symposium (IGARSS), 2012 IEEE International. 2012, pp. 3728–3731
  O. D’Hondt, S. Guillaso, and O. Hellwich
  
• “Bilateral filtering of PolSAR data based on Riemannian metrics”. In: Geoscience and Remote Sensing Symposium (IGARSS), 2012 IEEE International. 2012, pp. 5153–5156
  O. D’Hondt, S. Guillaso, and O. Hellwich
  
• “Extraction of points of interest from SAR tomograms”. In: Geoscience and Remote Sensing Symposium (IGARSS), 2012 IEEE International. 2012, pp. 471–474
  S. Guillaso, O. D’Hondt, and O. Hellwich
  
• “Complementarity of SAR Polarimetry and Tomography for Building Detection and Characterization”. In: Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci. Vol. XL-1/W1. 2013
  O. D’Hondt, S. Guillaso, and O. Hellwich
  
• “Geometric primitive extraction for 3D reconstruction of urban areas from tomographic SAR data”. In: Urban Remote Sensing Event (JURSE), 2013 Joint. 2013, pp. 206–209
  O. D’Hondt, S. Guillaso, and O. Hellwich
  
• “Iterative Bilateral Filtering of Polarimetric SAR Data”. In: Selected Topics in Applied Earth Observations and Remote Sensing, IEEE Journal of 6.3 (2013), pp. 1628– 1639. issn: 1939-1404
  O. D’Hondt, S. Guillaso, and O. Hellwich
  
• “Risk based parameter selection for polarimetric SAR speckle reduction”. In: Geoscience and Remote Sensing Symposium (IGARSS), 2014 IEEE International. 2014, pp. 4548–4551
  O. D’Hondt, S. Guillaso, and O. Hellwich
  (See online at https://doi.org/10.1109/IGARSS.2014.6947504)
• “Urban scene reconstruction from a reduced number of tomographic SAR data”. In: Geoscience and Remote Sensing Symposium (IGARSS), 2014 IEEE International. 2014, pp. 3164–3167
  S. Guillaso, O. D’Hondt, and O. Hellwich
  (See online at https://doi.org/10.1109/IGARSS.2014.6947149)