model-based polarimetric decomposition using polinsar coherence_v11(fileminimizer).pptx
TRANSCRIPT
Si-Wei Chen, Motoyuki Sato
Tohoku University, Japan
Model-based Polarimetric
Decomposition using PolInSAR
Coherence
2
Outline Introduction
– Current model-based decompositions
– Limitations
PolInSAR Coherence
– Estimation and optimization
Proposed Decomposition
– Adaptive volume scattering model
Comparative experiments
Conclusions
﹢ ﹢ ﹢...=C
Introduction
Polarimetric SAR (PolSAR)– Full polarimetric information
– Covariance matrix
Model-based decomposition– Better understanding the scattering mechanisms
3
2
2
2
2
2 2 2
2
HH HH HV HH VV
HH HV HV HV VV
HH VV HV VV VV
S S S S S
C S S S S S
S S S S S
(PolSARpro tutorials)
ALOS/PALSAR
Pre-event After-event
Optical Image
Decomposition
East Japan earthquake and tsunami
Ps Pv
Pd
Color-code
Double Bounce Volume Scattering Single Bounce
Model-based decomposition
Limitations
4
2
1 0
0 0 0
0
dbl dC f2
1 0
0 0 0
0
odd sC f
1 0 1 3
0 2 3 0
1 3 0 1
vol vC f
Freeman-Durden decomposition (1998)
Helix component
Nonnegative eigenvalues
Deorientation
Reflection symmetry assumption
Negative power
Scattering mechanism ambiguity
Inadaptive
(A. Freeman, Y. Yamaguchi, W.M. Boerner, J. J. Van Zyl, J.S. Lee, Y. Q. Jin, M.
Neumann, M. Arii and et al.)
General volume model
Single BounceDouble Bounce Volume Scattering
Improvements
Scattering mechanism ambiguity
General representation of the volume scattering
model
5
After
Deorientation19.48 13.52 16.75 12.79 18.81 18.53 17.56
223vP C33C 22C 224vP C 22
15
4vP CSPAN 11C
0
0 0
0
vol v
a d
C f b
d c
224vP C
22 224 15 4v vP C or P C
– For Freeman-Durden
– For Yamaguchi
Table I Averaged Backscattered Power (In dB)
221v v
a cP a b c f C
b223vP C
Decomposed volume scattering power
A
Pauli Image
Skew-oriented
building
Possible reasons and countermeasures Possible reasons
0
0 0
0
vol v
a d
C f b
d c
CountermeasuresAdaptive volume scattering model
Indirect modification of double- and single-bounce models
Balance the inputs and outputs
Utilization of both Polarimetric and
Interferometric information! 6
2
1 0
0 0 0
0
dbl dC f2
1 0
0 0 0
0
odd sC f
Double Bounce
2
22 2 HVC SOnly Volume scattering
Volume Scattering Single Bounce
Cross-
PolarizationTerrain slopes, oblique buildings
7
Outline Introduction
– Current model-based decompositions
– Limitations
PolInSAR Coherence
– Estimation and optimization
Proposed Decomposition
– Adaptive volume scattering model
Comparative experiments
Conclusions
PolInSAR coherence
Polarimetric SAR interferometry (PolInSAR)– Combination of PolSAR and InSAR
– Covariance matrix
Coherence magnitude
Optimization
8
1 12 2
1 2
1 11 1 2 22 2
ˆ( , ) ,
H
H HC C
11 12
6
12 22
H
CC
C
Polarimetric
Dependence
_1 _ 2 _ 3Opt Opt Opt1 2,
1 2
max
. . : 1s t
PolInSAR
(T. Xiong)
0 1
(K. Papathanassiou et al. )
Decorrelation sources– Signal-to-noise decorrelation
– Baseline decorrelation
– Processing decorrelation
9
sj
SNR temporal proc baseline volumee
NOTE: For manmade target
For forest
1, 1temporal volume
1, 1temporal volume
– Temporal decorrelation
– Volume decorrelation
– …
Sensitive to diverse terrains
PolInSAR coherence
Potentially, the volume scattering can be modeled from it!
Close relationship to forest structures
PolInSAR coherence:
10
PolInSAR Coherence
Optical image Optimal 1
Optimal 2 Optimal 3
11
PolInSAR Coherence
Optimal 1
Optimal 2 Optimal 3
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.01
0.02
0.03
0.04
0.05
0.06 Optimal 1
Optimal 2
Optimal 3
12
Outline Introduction
– Current model-based decompositions
– Limitations
PolInSAR Coherence
– Estimation and optimization
Proposed Decomposition
– Adaptive volume scattering model
Comparative experiments
Conclusions
Proposed decomposition
Adaptive volume scattering model
13
_
_
_
_
0
0 01
0
HH HH Opt n
Opt n
vol v
Opt n
Opt n VV VV
C f
1 0 1 3
0 2 3 0
1 3 0 1
vol vC f
Use Freeman-Durden model
1 0
0 1 0
0 1
vol vC f
(A. Freeman,2007)
_
_
2,
31
Opt n
Opt n
_
2
5Opt nIf
Modeled with PolInSAR coherence
Where, is adjust to the spatial and temporal baseline parameters.
Model compatibility
14
Model Parameters
_
_
_
_
0
0 01
0
HH HH Opt n
Opt n
vol v
Opt n
Opt n VV VV
C f
– More uniform distribution
– More sensitive for diverse
terrains
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.01
0.02
0.03
0.04
0.05
0.06 Optimal 1
Optimal 2
Optimal 3
Principles for the choice of
are: _Opt n
Two unknowns:
_ ,Opt n
_ _ 3Opt n Opt
15
Model Parameters
_
_
_
_
0
0 01
0
HH HH Opt n
Opt n
vol v
Opt n
Opt n VV VV
C f
_
22 11 33
0.5
1 ,
Opt n
vol vol volC C C
Indirect modification
of double and single
bounce scattering
models!
NOTE:
_
22
_
01
Opt n
vol
Opt n
C _1 max
0 1
Opt n
0.2 0.4 0.5 0.6 0.80
0.67
1
2
3
4
0.6
0.7
0.8
0.9
= 1
Opt_n
Cv
ol2
2
1For
2 2 22 ,HV HH VVS S S
Decomposition
Flowchart
Double & single bounce
model
– Indirect modification
Adaptive decomposition
– Pixel by pixel
– .
16
PolInSAR covariance
matrix
PolSAR covariance
matrix
_, ,HH VV Opt n
6C CPO angle
C Deorientation C
1 0 1 3
0 2 3 0
1 3 0 1
vol vC f
_
_
_
_
0
0 01
0
HH HH Opt n
Opt n
vol v
Opt n
Opt n VV VV
C f
Non-negative eigenvalues constraintremaider VolC C C
13Re 0C
1
13Re 0C
1
, ,d sf f , ,d sf f
,d sP P ,d sP P
, ,v d sP P P
v d sSPAN P P P
Yes No
Yes No
Double dominant Odd dominant
11 22 33SPAN C C C
_ 2 5Opt n
13Re 0C
vf
22vP C
2
22 2 HVC S
Volume scattering
Double bounce
Single bounce
17
Outline Introduction
– Current model-based decompositions
– Limitations
PolInSAR Coherence
– Estimation and optimization
Proposed Decomposition
– Adaptive volume scattering model
Comparative experiments
Conclusions
18
Experiment-I
E-SAR PolInSAR data
– Test site: Oberpfaffenhofen, Germany
– L-band
– Data size : 1300×1200
Azi
mu
th
Range
I
II
Optical image PolInSAR coherence RGB image
HH, HV, VV
Master track Pauli image
HH-VV, HV, HH+VV
E-SAR(PolSARpro tutorials)
19
Decomposition _ After deorientation
Full scene
Freeman-Durden Yamaguchi Proposed
Forest region
Freeman-Durden Yamaguchi Proposed
Ps Pv
Pd
Color-code
20
Skew-oriented built-up region
Freeman-Durden Yamaguchi Proposed
21
E-SAR PolInSAR dataBioSAR-2008 campaign L band Repeat-pass dataset
Spatial Baseline: 30 m Temporal baseline: 110min
Data size : 1496×840
Pauli Image Coherence RGB Image
Experiment - II
Optical Image
Mar. 2008 Jan. 2009 Oct. 2008
Azi
mu
th
Range
VV, HV, HHHH-VV, HV, HH+VV
Logged after the BioSAR 2008
22
Decomposition _ After deorientation
Freeman-Durden Yamaguchi ProposedOptical Image
More sensitive and better fit for diverse forest terrains!
Ps Pv
Pd
Color-code
23
Outline Introduction
– Current model-based decompositions
– Limitations
PolInSAR Coherence
– Estimation and optimization
Proposed Decomposition
– Adaptive volume scattering model
Comparative experiments
Conclusions
24
Conclusions
Adaptive volume scattering model
– Using PolInSAR coherence
– Better fit for different terrains
– Indirect modification of double- and single-bounce
scattering models
Adaptive decomposition
– Fully usage of the information
– Successfully discriminate the skew-oriented
buildings as manmade structures
– Overcome the scattering mechanism ambiguity
– Sensitive to diverse forest terrains
26
General representation of the volume scattering
model
Limitation of current model
Before
Deorientation19.48 14.10 15.26 14.69 20.71 20.43 19.47
After
Deorientation19.48 13.52 16.75 12.79 18.81 18.53 17.56
223vP C33C 22C 224vP C 22
15
4vP CSPAN 11C
0
0 0
0
vol v
a d
C f b
d c
224vP C
22 224 15 4v vP C or P C
– For Freeman-Durden
– For Yamaguchi
Table I Averaged Backscattering Power (In dB)
221v v
a cP a b c f C
b223vP C
Decomposed volume scattering power
A
Pauli Image
Skew-oriented
building
27
Optical image HH-HH
VV-VV HV-HV
PolInSAR Coherence
28
HH-HH
VV-VV HV-HV
PolInSAR Coherence
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.01
0.02
0.03
0.04
0.05
0.06
VV-VV
HV-HV
HH-HH
0.2 0.4 0.5 0.6 0.80
0.67
1
2
3
4
0.6
0.7
0.8
0.9
= 1
Opt_n
Cv
ol2
2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.05
0.1 Forest Region
Built-up Region
29
Model Parameters
Optimal 3 CoherenceOptical Image
_
_
_
_
0
0 01
0
HH HH Opt n
Opt n
vol v
Opt n
Opt n VV VV
C f
_
22
_
01
Opt n
vol
Opt n
C _1 max
0 1
Opt n
30
Skew-oriented built-up region
Freeman-Durden Yamaguchi Proposed
MethodBuilt-up area Forest area
Freeman-Durden 20 32 48 9 88 3
Yamaguchi 22 25 53 7 82 11
Proposed 29 8 63 13 81 6
sPdP vP sPdP vP
Table II Scattering Power Contribution (%)
Scattering power
31
Skew-oriented built-up region
Freeman-Durden Yamaguchi Proposed
32
Skew-oriented built-up region
Freeman-Durden Yamaguchi Proposed
33
Skew-oriented built-up region
Freeman-Durden Yamaguchi Proposed
Optimal 1
PolInSAR Coherence
Optimal 2 Optimal 3
Optimal 1
PolInSAR Coherence
Optimal 2 Optimal 3
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.01
0.02
0.03
0.04
0.05
Optimal 1
Optimal 2
Optimal 3
Optical Image Freeman-Durden Yamaguchi Proposed
Decomposition _ Volume scattering contribution
37
ALOS/PALSAR datasetsALOS/PALSAR
2007-4-02 2007-05-18
Pauli ImageOptical Image
Azi
mu
th
Range
Spatial baseline: 299m
Temporal baseline: 46 days
38
PolInSAR coherence _ H-V
HH-HH VV-VV HV-HV
39
PolInSAR coherence _ Optimal
Opt 1 Opt 2 Opt 3
40
PolInSAR coherence _ Histogram
41
Decomposition _ After deorientation
Freeman-Durden Yamaguchi Proposed
Ps Pv
Pd
Color-code
42
Built-up region - I
Optical image Freeman-Durden
ProposedYamaguchi
Ps Pv
Pd
Color-code
43
Built-up region - II
Optical image Freeman-Durden
ProposedYamaguchi
Ps Pv
Pd
Color-code