RADARSAT-2 and Polarmetric Applications Ian Cumming, UBC (

RADARSAT-2
and Polarmetric Applications
Ian Cumming, UBC
(ianc@ece.ubc.ca)
IET Radar Conference
Guilin, China, April 20-22, 2009
(slides 2-26 courtesy of MDA)
RADARSAT-2 Overview
MDA is the owner and operator of RADARSAT-2 and holds
the worldwide distribution rights for all products.
Launched in Dec. 2007 with a multi-mode, C-Band SAR
Mission duration: 7 years
Data continuity from RADARSAT-1
– all RADARSAT-1 imaging modes supported
– plus many additional capabilities
Co-funded by the Canadian Space Agency (CSA) and MDA
1
RADARSAT-2 – Features and Benefits
Offers a range of resolutions, swaths, and incidence angles
Finer resolution than RADARSAT-1 and other SAR satellites
Increased geometric accuracy
Operational multi-polarization modes
12-24 hour routine and 4-12 hour emergency programming
Fast processing and delivery
Solid-state recorders
Global Positioning System (GPS)
Reception at Beijing Ground Station since 2008
2
RADARSAT-2 Beam Modes
3
4
RADARSAT-2 Imaging Modes
5
RADARSAT-2 Polarization Diversity
RADARSAT-2 supports a variety of polarization modes that dramatically increase the
information content per pixel. The polarization modes are…
Selective Single
Polarization
HH or VV
or HV
– -22 dB NESZ (nominal)
HH or VV
Selective Dual
Polarization
– -23 dB NESZ (nominal)
HV or VH
+ Interchannel
Phases
Polarimetry (Quad Pol)
– -30 dB NESZ (nominal)
– relative phase error 5° HH
VV
HV
6
VH
Enables better
discrimination and
recognition of objects
on the ground and
improved
classification
capability
-- complementing
high-res optical
sensors
Provides for unsupervised
classification and much
better:
(1) target identification,
(2) change detection, and
(3) land cover type (surface
moisture, roughness,
vegetation cover)
Wide Beam Mode
Dual Pol Acquisition
(only VV channel shown)
Straight of Gibraltar
(Europe-Africa)
February 5, 2008
Wide 1 used (small
incidence angle)
Ocean features visible
with the HH co pol
channel, yet ships still
visible when ocean
clutter is low
7
Wide Beam Mode
Dual Pol Acquisition
(only VH channel shown)
Straight of Gibraltar
(Europe-Africa)
February 5, 2008
Wide 1 used (small
incidence angle)
Good ship detection with
the VH single cross pol
beam mode.
(1 bit BAQ)
8
ScanSAR Narrow
Dual Pol Acquisition
(only HH channel shown)
Gulf of St. Lawrence,
Canada, 2008
Circles indicate ships.
Similar to RADARSAT-1
ScanSAR.
Ocean surface features
more visible.
9
(1 bit BAQ)
ScanSAR Narrow
Dual Pol Acquisition
(only HV channel shown)
Gulf of St. Lawrence
Canada, 2008
Circles indicate ships.
New CROSS-POL channel
on RADARSAT-2.
HV suppresses ocean
clutter
10
(1 bit BAQ)
Fine Quad-Pol
San Francisco Bay, USA
April 2008
(single channels shown)
VV polarization provides best
ocean surface information.
VV
HV polarization
provides good discrimination
between surface
(ocean) and volume (land)
scattering,
and gives better ocean-target
contrast (ship in lower left).
VV and HV images are a subset of the
Fine Quad-Pol image
11
Google Earth ® image
(not coincident with
RADARSAT-2 acquisition time)
HV
Maritime Surveillance: Polarimetric Target Analysis
© 2 000
C a n a d ia n S p a c e A g e n c y , C a n a d ia n I c e S e r v ic e
H ib e r n ia
Cameron Decomposition
Ship: Dominant Scattering Mechanisms are:
Trihedral, Cylinder, Dipole
P r o d u c t io n
P la tfo rm
Iceberg: Dominant Scattering Mechanisms are:
Trihedral, Cylinder, (Dipole weaker)
Photo
12
Individual components
RADARSAT-2
Quad Pol
Vancouver
Airport
Trihedral
Polarimetric scattering
analysis:
Cylinder
Dipole
Narrow Diplane
Dihedral
Quarter Wave
13
Cameron
Decomposition
gives a “visual”
interpretation of the
types of scatterers
Polarimetric - Biomass Estimation
HH: Scattering is
from the canopy
HH
HV: Penetrates
Further
HV
VV
VV: Most Penetration
Ground
Return
Difference in Scattering Provides
estimate of Biomass and Height
14
Fine Quad-Pol
(SHH-SVV) (SHV+SVH)
(SHH+SVV)
(RGB composite, with
Pauli components)
Golden Gate Bridge &
San Francisco, USA
April 2008
RADARSAT-2 Image, Pauli color coded
NASA/JPL AIRSAR image of San Francisco (left)
With Freeman-Durden decomposition
15
Sinclair color coding – R (HH) + G (HV) + B (VV)
Fine Quad-Pol
Golden Gate Bridge &
San Francisco, USA
April 2008
28 nominal incidence angle
Most vegetation shows
green, as HV scattering
dominates.
Water shows blue, as VV is
the strongest.
Over the built up areas, the
dominant colors are white
and red. White pixels
correspond to equal
amplitude over all polarimetric
channels, whereas red
indicates that the phase
argument of HHVV* is close
to π, caused by a double
bounce reflection
VV
HV
HH
16
RADARSAT-2 Agricultural Example
Fine Quad-Pol
HH, HV and VV
(RGB color composite)
Brazil
February 19, 2008
Sugarcane fields
Composite image is
displayed in a red-greenblue colour scheme
(closest to natural colors)
17
Polarimetric Scene Analysis
Rough 91%
Double-bounce 3%
Volume 6%
HH
Delta, BC, Fine Quad Pol 15m res
May 6, 2008
3 km x 3 km subimage
18
Rough 56%
Double-bounce 24%
Volume 20%
RADARSAT-2 Polarization Diversity: Freeman-Durden decomposition
19
Quad Pol Applications
Derived Surface
Roughness Map
(RMS Height in cm)
Smooth
Rough
0.0 2.6 3.1 3.5 3.8 4.2 4.6 5.1 6.1
Relative Soil Moisture
Dielectric Constant)
Dryer
.00
20
Wetter
.35
.47
.55
.63
.71
.80
Tandem Demo DLR, July 23, 2008
TDM-PM-52- 7133 Issue/Rev: 0/1
.91
1.03
1.20
1.52
Disclosure restricted
as noted on the cover page.
Page 20
Improvements to Magnitude Change Detection:
Using Finer Resolution
UltraFine mode provides greater detail and context for change detection
Change detection based on Fine
mode data (~9 m resolution)
Change detection based on UltraFine
mode data (~3 m resolution)
Simulations based on Convair-580 airborne SAR data acquired over Vancouver airport
21
Improvements to Magnitude Change Detection:
Using Selectable Polarization
Selectable single- and dual-polarization provides a more robust capability for
detecting changes whose visibility is polarization dependent
Aircraft on
runway
Change Detection at HH
polarization
Change Detection at HV
polarization
RADARSAT-2 Ultrafine mode simulations from Convair-580 airborne SAR data
22
Polarimetric Change Detection and Change Classification
Polarimetry not only provides a novel capability for more comprehensive and
robust change detection, but also the potential for change classification
Repeatpass
images
Detail images
Quantify
Change
Change
Image
Detection
Change
Detection
Map
23
Image A / Image B
Change Image
Radiometric change
Pure polarimetric change
Data Fusion: Polarimetric and Higher Resolution Single-pol Data
Interpretation of polarimetric data and polarimetric change detection
will benefit from fusion with higher resolution single-polarization data
RADARSAT-2 Quad-Fine
polarimetric image
(~9m resolution,
-angle representation)
24
RADARSAT-2 Ultra-Fine
single-polarization image
(~3m resolution, HH)
Result of sharpening
polarimetric data with a
higher resolution image
Improvements to Coherent Change Detection
RADARSAT-2 provides substantial improvements to CCD
Increased resolution => reduces blurring inherent to spatial coherence estimation
Selectable polarization => coherence maps with higher SNR contrast
Improved orbit knowledge: better geocoding of coherence maps with other imagery,
e.g. optical high-resolution, ascending/descending SAR)
Better orbit control => smaller baselines allow to minimize spatial/volume
decorrelation and the topographic bias
25
RADARSAT-1
RADARSAT-2
Application
improvements
Orbit
knowledge
50 m (1 SA
off, postprocessed)
0.2 m (1 SA off,
post-processed)
Provides improved:
Registration accuracy
Baseline estimation
Reduction of artefacts
Orbit
control
2 km pipe
1 km pipe
Provides more suitable
baselines for CCD
Improvements to Persistent Scatter Interferometry
RADARSAT-2 will provide for more reliable and quicker PSINSAR analysis
Persistent scatterer interferometry:
– Uses image stacks to measure surface displacement to millimeter precision
– Applications: detecting subsidence, tunnelling and tectonic activity
PSINSAR will benefit from improvements in orbit knowledge and control of RADARSAT-2
RADARSAT-2 polarimetry also offers potential improvements to PSINSAR:
– Polarimetry may shorten the time required to acquire a reliable PSINSAR stack, for RADARSAT-1
this is currently 1-year (15 images).
– Polarimetry provides additional tools for testing the stability of persistent scatterers and classifying
them
Persistent scatterers detected
in different polarizations
26
A new object-oriented classification method
for POLSAR data
A two-step technique for classification of polarimetric SAR data is proposed: [1]
– We try to use the visual information content that humans utilize when they perform
manual segmentation
– Based on “Spectral Graph Partitioning” that was shown to perform well on grouping
problems in computer vision.
– “Contour” and “Spatial Proximity” information is used for segmentation
– Classification is based on pair-wise similarity of mean coherency matrices of segments

Results are given on subsets of CV-580 (C-band) and AIRSAR (L-band) data
Perceptually plausible results: more homogenous, agree with the reference
(i.e., manual) segmentation
Resulting segmentation is cleaner and classification is better than Wishart
This scheme is flexible: allows further improvement using additional information
[1] K. Ersahin, I. Cumming and R. Ward, “Segmentation and Classification of Polarimetric SAR Data
Using Spectral Graph Partitioning”, Trans. of Geoscience and Remote Sensing (Under Review)
27
Proposed Scheme
PolSAR Data
Classification
Segmentation
Multi-looking

Perform multi-looking on SLC data set

Form affinity matrix W for each data channel
Proximity
To account for proximity in the image plane, calculate affinities
only within a neighborhood.
Contour
Information
Contour information is measured using Orientation Energy (OE)

Perform the Spectral Graph Partitioning (SGP) using the
Multiclass Spectral Clustering (MSC) algorithm.

Form pairwise similarity matrix W and perform SGP
SGP
Similarity of
Coherency
Matrices
SGP
Segments obtained from the previous step are used to estimate
mean coherency matrices for segments.
Symmetric Revised Wishart distance (Anfinsen et al.) is used to
group segments into classes.
28
CV-580 Data Set: Westham Island, BC, Canada
Potatoes
Hay
Barley -1
Pumpkin
Bare Soil
Barley – 2
Turnip
Strawberry
29
Westham Island Results: CV-580 data
RGB (HH-HV-VV)
Ground Truth
Wishart Result
Proposed SGP
Technique
Wishart: 81.6 %
SGP: 86.4 %
Wishart: 74.1 %
SGP: 93.0 %
Wishart: 63.1 %
SGP: 81.0 %
30
Flevoland Results: AIRSAR data
RGB (HH-HV-VV)
Wishart Result ( 74.1 % )
Ground Truth
Proposed SGP Technique ( 81.2 % )
31
Conclusions
RADARSAT-2 is working well, and is available for data takes
around the world
– all RADARSAT-1 imaging modes supported
– plus many additional capabilities
Several multi-polarization modes are available to enhance
image interpretability
– dual polarization modes offer wide swaths, with improved image
interpretation over single-channel modes
– quad-polarization mode offers superior image classification, at
the expense of narrower swath widths
32