Header

UZH-Logo

Maintenance Infos

Deriving high resolution 3D information using multibaseline airborne SAR interferometry


Magnard, Christophe. Deriving high resolution 3D information using multibaseline airborne SAR interferometry. 2016, University of Zurich, Faculty of Science.

Abstract

Synthetic aperture radar interferometry (InSAR) is a remote sensing technology that can be used to generate topography measurements such as digital elevation models (DEMs). The measurements are characterized by properties that highly depend on the system parameters. These properties include backscatter characteristics, penetration into vegetation, spatial resolution, geolocation accuracy, vertical noise, and coherence.
InSAR sensors have been used operationally on air- and spaceborne platforms to generate moderate resolution DEMs. They are able to cover large areas with a small number of acquisitions, with independence from weather and daylight conditions. Higher, submeter resolution topography measurements are typically achieved using airborne laser scanning (ALS) technology. While now a mature and widely-used technology, ALS uses narrow swaths for high resolution data acquisitions and therefore requires large numbers of acquisitions. InSAR has the potential of achieving high resolution 3D measurements using the appropriate system parameters, and therefore could combine high resolution data acquisition with a large coverage, reducing costs.
Data from an experimental airborne millimeter wave multibaseline InSAR system, the MEMPHIS system, were studied in this work. The datasets yielded decimeter resolution SAR images. The sensor uses the Ka-band wavelength that penetrates only marginally into vegetation, and multiple receiving antennas to combine straightforward phase unwrapping with reduced vertical noise. These unique properties enable the generation of high resolution three dimensional data, with accuracies that approach those of ALS data.
To fully exploit the provided raw data, they first have to be focused into high resolution SAR images. Subsequent interferometric processing yields the topographic information. The work presented here investigated the prerequisites, such as the required navigation data quality and appropriate SAR focusing algorithms, necessary to generate well-focused images, accurate geo- location, and trustworthy interferometric phase values.
SAR focusing and interferometric processing algorithms able to generate highly accurate point clouds or digital surface models (DSMs) are proposed. First a coarse-to-fine (C2F) interferometric phase estimation method is introduced, where interferograms generated using the shorter base- lines are only used to help unwrap the interferogram generated using the longest available base- line. A maximum likelihood (ML) phase estimation method was also investigated, taking into account information from all receiving antennas to retrieve a better phase estimation. Signatures of corner reflectors were analyzed in the generated SAR images: measured resolutions yielded values close to the theoretical expectations. The planimetric geolocation accuracy was typically better than 0.1 m, validating the SAR system and focusing algorithm. InSAR DSMs were com- pared to ALS-based models to validate their absolute vertical accuracy; in grassland areas, the height difference between the ~2 m-resolution InSAR DSMs and the reference ALS models was typically 0 ± 0.25 m. The performance of the ML phase estimation was compared to results based on the C2F algorithm, with the ML method consistently delivering higher accuracies: the noise level using the ML approach was slightly but steadily lower than the noise level obtained using the C2F method.
The potential of InSAR-based point clouds is demonstrated in an application: the point cloud of a forest canopy was generated from multi-aspect multibaseline InSAR data and compared to equivalent products generated using ALS and stereo-photogrammetry techniques. Through a seg- mentation of the point cloud, single trees were detected and their position, height, and crown diameter estimated. These estimates were compared to reference forestry data. The InSAR, ALS, and photogrammetry-based point clouds all showed similar geolocation accuracies, with 0.2 – 0.3 m relative shifts. A much more limited penetration into the canopy was observed for both the InSAR and photogrammetry derived point clouds as compared to ALS. Canopy height models agreed very well with each other, with the InSAR height ~1 m lower than those derived from the other point clouds. Most of the large trees were accurately detected, as well as approxi- mately half of the smaller trees, with a localization accuracy typically better than 1 m. Asides from a slight underestimation, the tree heights agreed well with the reference data, and the esti- mation of the crown diameter was accurate in the mean. Results were more accurate for conifers than for broad-leaf trees. All these results are in line with similar studies that tested ALS data. They validate millimeter wave multibaseline InSAR-data as a reliable alternative for forest mon- itoring in comparison to other remote sensing techniques such as ALS and stereo-photogramme- try.
In the final chapter, a synthesis of the main findings is presented. The successful use of milli- meter wave multibaseline SAR interferometry to carry out reliable and accurate high resolution topography measurements is highlighted, including its application in monitoring forested ecosys- tems. Limitations are identified: they range from motion compensation errors, to hardware-related severe range sidelobes when illuminating targets with intense backscattering, and to limited pen- etration into the forest canopy reducing detection of understory trees. Further improvements to the data processing are suggested: a reduction of the range sidelobes using multi-step adaptive pulse compression, an elaborate sample selection for the phase estimation, and the use of SAR tomography. Finally, several potential applications are proposed, such as monitoring and analysis of man-made objects, land cover classification using a combination of InSAR and polarimetric data, or change detection. In the latter case, the inherent 3D geolocation of the points removes the need for a precise external height model, and shadowed areas can be filtered using a coherence threshold, reducing the number of false alarms. Through its thorough validation, this work paves the way toward a more operational airborne or possibly spaceborne system that could combine high resolution topography measurements with a wide coverage.

Abstract

Synthetic aperture radar interferometry (InSAR) is a remote sensing technology that can be used to generate topography measurements such as digital elevation models (DEMs). The measurements are characterized by properties that highly depend on the system parameters. These properties include backscatter characteristics, penetration into vegetation, spatial resolution, geolocation accuracy, vertical noise, and coherence.
InSAR sensors have been used operationally on air- and spaceborne platforms to generate moderate resolution DEMs. They are able to cover large areas with a small number of acquisitions, with independence from weather and daylight conditions. Higher, submeter resolution topography measurements are typically achieved using airborne laser scanning (ALS) technology. While now a mature and widely-used technology, ALS uses narrow swaths for high resolution data acquisitions and therefore requires large numbers of acquisitions. InSAR has the potential of achieving high resolution 3D measurements using the appropriate system parameters, and therefore could combine high resolution data acquisition with a large coverage, reducing costs.
Data from an experimental airborne millimeter wave multibaseline InSAR system, the MEMPHIS system, were studied in this work. The datasets yielded decimeter resolution SAR images. The sensor uses the Ka-band wavelength that penetrates only marginally into vegetation, and multiple receiving antennas to combine straightforward phase unwrapping with reduced vertical noise. These unique properties enable the generation of high resolution three dimensional data, with accuracies that approach those of ALS data.
To fully exploit the provided raw data, they first have to be focused into high resolution SAR images. Subsequent interferometric processing yields the topographic information. The work presented here investigated the prerequisites, such as the required navigation data quality and appropriate SAR focusing algorithms, necessary to generate well-focused images, accurate geo- location, and trustworthy interferometric phase values.
SAR focusing and interferometric processing algorithms able to generate highly accurate point clouds or digital surface models (DSMs) are proposed. First a coarse-to-fine (C2F) interferometric phase estimation method is introduced, where interferograms generated using the shorter base- lines are only used to help unwrap the interferogram generated using the longest available base- line. A maximum likelihood (ML) phase estimation method was also investigated, taking into account information from all receiving antennas to retrieve a better phase estimation. Signatures of corner reflectors were analyzed in the generated SAR images: measured resolutions yielded values close to the theoretical expectations. The planimetric geolocation accuracy was typically better than 0.1 m, validating the SAR system and focusing algorithm. InSAR DSMs were com- pared to ALS-based models to validate their absolute vertical accuracy; in grassland areas, the height difference between the ~2 m-resolution InSAR DSMs and the reference ALS models was typically 0 ± 0.25 m. The performance of the ML phase estimation was compared to results based on the C2F algorithm, with the ML method consistently delivering higher accuracies: the noise level using the ML approach was slightly but steadily lower than the noise level obtained using the C2F method.
The potential of InSAR-based point clouds is demonstrated in an application: the point cloud of a forest canopy was generated from multi-aspect multibaseline InSAR data and compared to equivalent products generated using ALS and stereo-photogrammetry techniques. Through a seg- mentation of the point cloud, single trees were detected and their position, height, and crown diameter estimated. These estimates were compared to reference forestry data. The InSAR, ALS, and photogrammetry-based point clouds all showed similar geolocation accuracies, with 0.2 – 0.3 m relative shifts. A much more limited penetration into the canopy was observed for both the InSAR and photogrammetry derived point clouds as compared to ALS. Canopy height models agreed very well with each other, with the InSAR height ~1 m lower than those derived from the other point clouds. Most of the large trees were accurately detected, as well as approxi- mately half of the smaller trees, with a localization accuracy typically better than 1 m. Asides from a slight underestimation, the tree heights agreed well with the reference data, and the esti- mation of the crown diameter was accurate in the mean. Results were more accurate for conifers than for broad-leaf trees. All these results are in line with similar studies that tested ALS data. They validate millimeter wave multibaseline InSAR-data as a reliable alternative for forest mon- itoring in comparison to other remote sensing techniques such as ALS and stereo-photogramme- try.
In the final chapter, a synthesis of the main findings is presented. The successful use of milli- meter wave multibaseline SAR interferometry to carry out reliable and accurate high resolution topography measurements is highlighted, including its application in monitoring forested ecosys- tems. Limitations are identified: they range from motion compensation errors, to hardware-related severe range sidelobes when illuminating targets with intense backscattering, and to limited pen- etration into the forest canopy reducing detection of understory trees. Further improvements to the data processing are suggested: a reduction of the range sidelobes using multi-step adaptive pulse compression, an elaborate sample selection for the phase estimation, and the use of SAR tomography. Finally, several potential applications are proposed, such as monitoring and analysis of man-made objects, land cover classification using a combination of InSAR and polarimetric data, or change detection. In the latter case, the inherent 3D geolocation of the points removes the need for a precise external height model, and shadowed areas can be filtered using a coherence threshold, reducing the number of false alarms. Through its thorough validation, this work paves the way toward a more operational airborne or possibly spaceborne system that could combine high resolution topography measurements with a wide coverage.

Statistics

Altmetrics

Downloads

82 downloads since deposited on 14 Feb 2017
60 downloads since 12 months
Detailed statistics

Additional indexing

Item Type:Dissertation
Referees:Schaepman Michael E, Meier Erich, Small David, Stilla Uwe
Communities & Collections:07 Faculty of Science > Institute of Geography
Dewey Decimal Classification:910 Geography & travel
Language:English
Date:2016
Deposited On:14 Feb 2017 14:01
Last Modified:02 Feb 2018 12:19
Number of Pages:126
ISBN:978-3-906894-00-3
Additional Information:Erschienen in Remote sensing series; 71
OA Status:Green
Free access at:Related URL. An embargo period may apply.
Related URLs:https://www.recherche-portal.ch/ZAD:default_scope:ebi01_prod010802325 (Library Catalogue)

Download

Download PDF  'Deriving high resolution 3D information using multibaseline airborne SAR interferometry'.
Preview
Content: Published Version
Filetype: PDF
Size: 31MB