6. We provide…
Powerful and scalable
image processing
solutions that let you
quickly and efficiently
produce information
products from any type
of imagery
SCALABLE TO WE ARE UNMATCHED
ANY SIZE SENSOR AUTOMATED
PROJECT AGNOSTIC WORKFLOWS
BUILDING
HIGH SPEED ADVANCED
SOLUTIONS
MULTI FOR RADAR
CPU / GPU CAPABILITY
30 YEARS
7. WHICH SOLUTION IS RIGHT FOR YOU?
$1M
$500
Price
($000’s)
$200
$10
10 GB 100 GB 500 GB 1 - 5TB 5 - 10TB
Page 7
8. PCI – SAR technology development
Canada has been an innovator in SAR data
acquisition and processing since the early
1980s – PCI has been involved since the
beginning
PCI Geomatics participated in GlobeSAR
program, delivered training and software
PCI Geomatics developed technology
through Canadian Government (SAR
Polarimetry Workstation)
PCI Geomatics works with multi-sensor SAR
imagery
Page 8
9. SAR Sensor Support
RADARSAT 1 & 2
TerraSAR-X
Cosmo-SkyMed,
UAVSAR
PALSAR
ASAR
ERS 1 & 2
Page 9
10. Generic SAR Capabilities
Support for Single, Dual, Quad, Data
Automatic Calibration*
Automatic Geocoding*
Speckle Filtering (many)
Statistical & Analysis Capabilities
Ortho-rectification, Integration and
Visualization with Optical Data
* If available
Page 10
12. Advantages for applications
Key Advantages of Commercial Radar Imagery
– Data collections are independent of lighting and cloud conditions
– Frequent imaging supports routine change detection
– Provides effective wide area (100 –500+ km swath) coverage
– A variety of information is contained in the return signal that can be
extracted
Key Maritime Missions:
– Large Area Maritime Domain Awareness
– Efficient Tasking of Patrol Assets
– Monitoring Port Activity
Key Terrestrial Missions:
– Change Detection
– Disaster Response
– DEM Generation
Page 12
15. 1. Amplitude Change Detection
Different sensors / beam modes /
resolutions can be used in combination
Revisit is more important in this case than
matching geometry
Presence / absence of features readily
observed
Page 15
19. 2. Coherent Change Detection
Measures phase differences in SAR signal
Geometry must be matching (repeat pass)
Multiple collections over same area from
different sensors/orbits can be combined
Page 19
22. Coherent Change Detection
Loss of Coherence
is indicated by
Dark Colour
Note:
Loss of Coherence for Trees
Page 22
23. Cross Sensor Change Detection
Sample CCD over Flevoland
TerraSAR-X and RADARSAT-2
acquisitions
Two sets of repeat pass collections
PCI Technology used to achieve high
cross-sensor image registration
Page 23
33. 3. Polarimetric Analysis and Change
Detection
Basics of Polarimetry
Polarimetric information for ship dectection
Page 33
34. Some Polarimetric Basics
V
For a single
polarization, the
return is
proportional to the
target cross
section.
H
For HH we would
get a return
indicated by red.
For VV it would be
blue.
So the amount of return we get depends on
target orientation and polarization
Page 34
35. Some Polarimetric Basics
V
For a single
polarization, the
return is
proportional to the
target cross
section.
For HH we would H
get a return
indicated by red.
For VV it would be
blue.
So the amount of return we get depends on
target orientation and polarization
Page 35
36. Some Polarimetric Basics
X Polarimetric radar
data provides full
scattering information
in the direction of the
line of sight
Y
Y X
We want to compare these targets.
Page 36
37. Some Polarimetric Basics
Polarimetric radar
data provides full
scattering information
in the direction of the
Y line of sight
Y H X
Y
X
H
We can do some fancy arithmetic and rotate the
scattering matrix until we get a maximum X and a
minimum Y.
Then we can compare their properties. Page 37
38. Non-polarimetric Parameters
Time 2001-02-30 12:34:56 GMT
Position: 12:01:21.58 N 34:14:43.37 W
Incidence Angle: 27.15°
Estimated Length: 226 m
Estimated Heading: 260°
Estimated Velocity: 9.70 m/s
Page 38
39. Polarimetric Processing Steps
Ingest Full Polarimetric Data
(Optionally) calibrate to σ
Apply multi-channel speckle filter
Decompose (Cloude-Pottier) image into (16) polarimetric classes
Iterate (3-5 times) to enhance classification and remove outliers
Exclude pixels from the largest class (which will be water)
Generate land mask *
Generate polarimetric parameters using FOCUS, SPW and SPTA
from remaining (non-masked) pixels
Page 39
41. Polarimetric Information
Maximum of the degree of polarization: 0.7916655
Minimum of the degree of polarization: 0.09595539
Maximum of the completely polarized component: 2.520944
Minimum of the completely polarized component: 0.2940039
Orientation of Maximum Polarisation 70
Ellipticity of Maximum Polarisation -5
Maximum of the completely unpolarized component: 2.769960
Minimum of the completely unpolarized component: 0.6619406
Maximum of the scattered intensity: 3.210612
LL
Minimum of the scattered intensity: 2.850842
Coefficient of Variation: 0.1160221
Fractional Power: 0.7920792
HH VV
Pedestal Height 1.318336
HH / HV Ratio 4.014223
HH / HV Correlation 0.2035844 RR
HH / VV Ratio 0.9518262
Page 41
HH / VV Correlation 0.3857002
42. Polarimetric Signature Information
V
LL
5° Ellipticity
70° Orientation
H
VV
Maximum Return
V
LL
RR
Secondary
HH Return Max
Return
H
- 20° Orientation
Strong Secondary Return
Page 42
RR
44. Polarimetric Decompositions
Cloude-Pottier
Target Average % High % Medium % Low
Entropy 0.8480822 2.253302 76.30148 21.44522
Anisotropy 0.5064220 55.63326 44.36674
Alpha Angle 43.200062 27.50583 30.53613 41.95804
Touzi (ICTD)
Target Tilt Dominant Eigen Symmetric Symmetric Helicity
Angle Value Scattering Type Scattering Type (Symmetry)
(deg) Magnitude Phase (deg)
-27.432373 0.5600992 10.467688 -50.483246 5.841676
van Zyl
% Unclassified % Odd % Even % Diffuse
1.892744 48.264984 23.343849 26.498423
Page 44
45. van Zyl Decomposition
Radar Measurement Physical Meaning
Odd Number Bounce Flat Surface
Even Number Bounce Superstructure
Diffuse Scattering Complex / Random
Page 45
46. Symmetric Scattering Decomposition
Trihedral
(odd number of bounces)
Cylinder
(weak return in one direction)
Dipole
(no return in one direction)
Quarter Wave
(delay in second direction)
Dihedral
(even number of bounces)
Narrow Dihedral
(with one direction attenuated)
Page 46
55. Stereo DEMs
All or
Maximum Overlap
Next Pair
Image match based upon Power
Linear or Decibels
Image A Image B
No
Overlap, Look Direction
Angular Difference Suitable Pair ?
Downsample Image A Downsample Image B
to User Specs to Epipolar Image A
Spacing Affects DEM
Detail Level Extract Window Area Extract Search Area
Ignore Background Find Common Points
(No Data) Pixels
Stereo Intersection
Store Elevation
No
Last Pair ?
Blend Overlap Areas
Last, Average, High Score Arbitrate Values
Write Failed Value where Fill Gaps/Holes
“gaps” remain
Remove “buildings “ * Write Final DEM
56. Suggestions for Selection of Stereo Pairs
Selection of Stereo Image Pairs
Candidate pair should have more than 50 % overlap
Candidate pair should have nominally the same resolution
Best results obtained from same-side (i.e., descending/descending
or ascending/ascending) image pairs
Candidate pair should have matching polarizations
Large incident angle (i.e., S7 ) are preferable (to minimize terrain
displacement effects)
The larger the difference between incident angles, the greater the
parallax in the stereo pair (recommend 5 - 25 angular difference)
Opposite-side (i.e., ascending/descending) image pairs only
recommended for very low relief areas; with similar tonal
characteristics
67. Wind Speed Analysis
Steps:
#1: Convert to calibrated data (SARINGEST)
#2: Boxcar Filter (19 x 19)
#3: Convert filtered HV data to decibel
#4: If HV data ( < -21 dB) apply Paris Vachon
algorithm to generate Windspeed in m/s.
Purple = 10 m/s to red = 16 m/s.
Page 67
71. Summary of PCI Capabilities
Software / scalable Experience/ know-how
Geomatica Radar Suite Dedicated development
www.pcigeomatics.com/sar team
Includes SPW and Target Senior SAR scientist on
Analysis
Ingest, correct Multi-sensor
staff
SAR data 30 years of experience
SAR training available
SAR for GXL
Custom implementation of
SAR analysis for large
volume processing
Page 71