Processing of MEMPHIS mmW multi-baseline InSAR data Christophe Magnard, Erich Meier and David Small Remote Sensing Laboratories Department of Geography University of Zurich, Switzerland www.rsl.ch Honolulu, July 30, 2010 Helmut Essen and Thorsten Brehm Fraunhofer-Institut für Hochfrequenzphysik und Radartechnik FHR Wachtberg, Germany www.fhr.fgan.de
17
Embed
FR2.L09 - PROCESSING OF MEMPHIS MILLIMETER WAVE MULTI-BASELINE INSAR DATA
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Processing of MEMPHIS mmWmulti-baseline InSAR data
Christophe Magnard, Erich Meierand David Small
Remote Sensing LaboratoriesDepartment of Geography
University of Zurich, Switzerlandwww.rsl.ch
Honolulu, July 30, 2010
Helmut Essen and Thorsten Brehm
Fraunhofer-Institut für Hochfrequenzphysik und Radartechnik FHRWachtberg, Germany
www.fhr.fgan.de
Overview
• MEMPHIS SAR system• Processing method
– Data focusing– Multi-baseline processing– Phase to digital surface model conversion– Correction of azimuth phase undulations
• Experimental results• Discussion
1/15
MEMPHIS SAR system (I)
• MEMPHIS (Multi-frequency Experimental Monopulse High-resolution Interferometric SAR) developed by Fraunhofer-FHR
• Multi-baseline cross-track interferometric system
• Longest baselines: 0.275 m for the 35 GHz antenna, 0.16 m for the 94 GHz antenna
Platform C-160 Transall
Available frequencies 35 and 94 GHz (Ka- and W- bands)
Bandwidth 800 MHz (stepped frequency chirp)
PRF 1500 Hz
Typical velocity 75 m/s
Typical near range distance 1.5 km
Flying altitude 300 – 1000 a.g.l.
Depression angle 20° – 35°
Typical scene dimension 600 m swath width x 3000 m azimuth length
MEMPHIS 35 GHz antenna
2/15
MEMPHIS SAR system (II)
• Experimental, removable system• Inclined depending on the data take
accuracy ~ a few centimeters• Kalman filtering used to combine
dGPS and INS data• Corner reflectors placed at all test
sites for data calibration
3/15
range
azi
mu
th
Data focusing
• The SAR data are processed and focused to obtain single look complex (SLC) images
• The same parameters are used for each receiver (horn)
• Typical procedure for high resolution processing with -k:
1. Extraction of the SAR raw data
2. Range compression of data from each stepped chirp (8 stepped chirps)
3. Stepped frequency chirp processing
4. First order motion compensation
5. Azimuth compression
Ka-band SAR image (resolution 0.2 m)
4/15
Multi-baseline processing (I)
• Two multi-baseline processing algorithms were tested:
1. Coarse to fine phase unwrapping process: shorter baselines used exclusively to assist in unwrapping of the longest-baseline interferogram1, 2
Interferograms are calculated for each possible combination of SLC data(i.e. for each possible baseline)
Clock analogy:– Interferogram with the smallest baseline (i.e. with the largest height
ambiguity) used as the reference unwrapped phase (≈ hour hand)– Interferograms with longer baselines are used to refine the unwrapped
phase (≈ minute, second hand)
1 H. Essen, T. Brehm, S. Boehmsdorff, U. Stilla, “Multibaseline Interferometric SAR at Millimeterwaves, Test of an Algorithm on Real Data and a Synthetic Scene”, Proceedings of SPIE, vol. 6746, Sept 2007.
2 C. Magnard, E. Meier, M. Rüegg, T. Brehm, H. Essen, “High Resolution Millimeter Wave SAR Interferometry”, Proceedings of the IEEE International Geoscience and Remote Sensing Symposium IGARSS , pp. 5061 – 5064, Barcelona, July 2007.
5/15
Multi-baseline processing (II)
2. Maximum Likelihood (ML)1:– Direct use of SLC data array– Find optimum interferometric phase using ML estimator
• In both cases, remaining height ambiguities (related to the smallest baseline, given significant altitude changes within the scene) can be unwrapped with a conventional phase unwrapping algorithm.
• We use the statistical-cost network-flow algorithm for phase unwrapping (SNAPHU2)
1 P. Lombardo, F. Lombardini, “Multi-baseline SAR Interferometry for Terrain Slope Adaptivity”, IEEE Proc. Nat. Radar Conf. 1997, pp. 196-201, New York, May 1997.
2 C.W. Chen, “Statistical-Cost Network-Flow Approaches to Two-Dimensional Phase Unwrapping for Radar Interferometry”, PhD Dissertation, Department of Electrical Engineering, Stanford University, 2001.
6/15
Phase to digital surface model conversion
• Conversion of range and azimuth positions and their phase differences into digital surface model (DSM) with help of linearized flight track and tie points.
• Data processed in a zero-Doppler geometry an appropriate coordinate system transformation enables simplifications
• Tie points used to determine constant phase offset const
• The geographical position and height of each point can then be computed• Equations and approach are derived from 1 and 2
• Regridding subsequently required to rasterize the DSM
1 P. Rosen, S. Hensley, I. Joughin, F. Li, S. Madsen, E. Rodriguez, R. Goldstein, “Synthetic Apertur Radar Interferometry”, Proceedings of the IEEE, vol. 88, no. 3, pp. 333-381, March 2000.
2 D. Small, C. Werner, D. Nüesch, “Baseline modelling for ERS-1 SAR interferometry”, Proceedings of the IEEE International Geoscience and Remote Sensing Symposium IGARSS, vol. 3, pp. 1204-1206, Tokyo, August 1993.
7/15
Correction of azimuth phase undulations (I)
• Aircraft attitude variations cause fluctuations in the baseline components Phase undulations in interferograms
• Two possible methods to handle this:
1. Use attitude data measured by aircraft INS system to correct baseline for each pixel position
– INS accuracy usually acceptable, but sampling rate insufficient Results are not accurate enough; method not applicable
2. Use interferometric data to correct azimuth phase undulations1, 2
1 P. Prats, J.J. Mallorqui, “Estimation of azimuth phase undulations with multisquint processing in airborne interferometric SAR images”, IEEE Transactions on Geoscience and Remote Sensing, vol. 41, no. 6, pp. 1530-1533, 2003.
2 P. Prats, A. Reigber, J.J. Mallorqui, “Interpolation-Free Coregistration and Phase-Correction of Airborne SAR Interferograms”, IEEE Geoscience and Remote Sensing Letters, vol. 1, no. 3, pp. 188 - 191, 07/2004.
8/15
Correction of azimuth phase undulations (II)
a. Azimuth focusing performed with two different Doppler centroids, use of a reduced azimuth bandwidth to avoid overlapping spectra
b. Interferograms generated for both Doppler centroids, followed by a flat Earth removal and phase unwrapping
c. Integration of the phase difference between the two interferograms
d. Small squint angle difference (narrow azimuth beam width) can cause large integration errors Linear correction of integrated phase correction using INS attitude data Kalman filtering to combine phase corrections calculated from INS and
InSAR data
e. Correction applied to range compressed data
9/15
Correction of azimuth phase undulations (III)
Interferogram without phase correction Interferogram with phase correction
range
azi
mu
th
range
azi
mu
th
+
+
0
-
-
10/15
Experimental results
• InSAR flight campaign in 2009 over test sites in Switzerland
• Results presented: highway roundabout near Zurich, Ka-band antenna
• Comparison of the masked InSAR DSM with a reference LIDAR DTM
• Important: measured heights not identical (different physical phenomena, different vegetation states) Histogram not a perfect normal
distribution
• For statistical information, negative side of the histogram is mirrored and standard deviation calculated: 0.672 m for this dataset.
• Similar results obtained with other Ka-band datasets
14/15
Discussion
• Main difficulty with MEMPHIS: producing precise and accurate products with imperfect navigation data and movable hardware components
• Methods to overcome uncertainties:
– Tiepoints for calibrating navigation data and calculating const
– Correction of azimuth phase undulations using spectral diversity(or multi-squint) method combined with attitude data
• Use of the ML algorithm more difficult than expected: measurement of smallest baselines associated with high uncertainty Difficult to assess if ML phase is meaningful or corrupted by measurement
uncertainties in smaller baselines Results presented obtained with “coarse to fine” method
• Results in good agreement with LIDAR height models; standard deviation of their filtered difference ~0.6 – 0.7 m