Visualization, Detection and Reconstruction of Trees in Fully Polarimetric Long Wavelength SAR Imagery

Kurzfassung

State of the art synthetic aperture radar (SAR) imaging and tomographic signal processing techniques allow for the fully three dimensional reconstruction of remotely sensed volumetric structures on or near the Earth's surface, such as snow and ice or forest. Recent studies have demonstrated that high resolution tomographic SAR reconstructions can be used to provide detailed characterizations of 3D structures such as individual buildings in urban environments [1] or individual trees within forested areas [2]. These studies were based on SAR data acquired at very short, i.e. sub-centimeter, wavelengths and highlighted the importance of multi-aspect data sets, in which the region of interest is imaged from several look-directions, in the extraction of 3D structures. The research objective of the present thesis is to determine whether the proposed techniques can be modi�ed and extended to polarimetric SAR data acquired at signi�cantly longer wavelengths such as L-Band (20cm wavelength). The emphasis of the investigation lies in the analysis of forested areas, where applications such as biomass estimation from polarimetric long-wavelength SAR data are a topic of intense research activity as evidenced, for example, by ESA's upcoming space borne BIOMASS mission and the recent airborne SAR activities in the context of the AFRISAR campaign carried out by ONERA and DLR. The investigation is based on very high resolution multi-circular SAR data acquired by DLR's F-SAR sensor. This innovative SAR imaging mode, in which the sensor acquires a SAR data set in a spiral ight trajectory above the test site, can be processed to obtain a high resolution, in the decimeter range, 3D reconstruction despite the long wavelengths. In contrast to the SAR data used in previous studies, the acquired multi-circular data are also fully polarimetric, such that di�erent types of physical scattering mechanism can be distinguished in each 3D resolution cell. Two central questions are to be answered with respect to the example application of segmenting and reconstructing individual trees in forested areas: � To what extent do approaches established for short-wavelength SAR data apply to long-wavelength SAR data, where the resolution of the 3D reconstruction is generally lower and the radar wave penetrates signi�cantly deeper into the structures of interest (i.e. the forest)? How can existing approaches be adapted and extended to improve performance for this type of SAR data? � How can existing approaches be extended to take advantage of the added information in multi-channel polarimetric SAR data? Can the segmentation algorithm be modi�ed to incorporate prior knowledge concerning the physical backscattering mechanisms that are typical of the objects of interest (e.g. volume scattering in the canopy and double-bounce re ection on the tree trunk)? As part of the investigation, informative 3D visualizations of the input SAR data as well as the segmentation results are to be developed. These are required to allow a visual interpretation of the input data as well as a qualitative evaluation of segmentation results.