SAR Imagery Algorithms Simulated data Real data Conclusion and Future Work Contribution of the polarimetric information in order to discriminate target from interferences subspaces. Application to FOPEN detection with SAR processing 1 F.Brigui a , L.Thirion-Lefevre b , G.Ginolhac c and P.Forster c a ISAE/University of Toulouse b SONDRA/SUPELEC c SATIE, Ens Cachan 1 Funded by the DGA 1/24 IGARSS 2011 July 2011
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
SAR Imagery AlgorithmsSimulated data
Real dataConclusion and Future Work
Contribution of the polarimetric information inorder to discriminate target from interferencessubspaces. Application to FOPEN detection
with SAR processing 1
F.Briguia, L.Thirion-Lefevreb, G.Ginolhacc and P.Forsterc
aISAE/University of Toulouse
bSONDRA/SUPELEC
cSATIE, Ens Cachan
1Funded by the DGA1/24 IGARSS 2011 July 2011
SAR Imagery AlgorithmsSimulated data
Real dataConclusion and Future Work
Context
Objective
Detection of a target embedded in a complex environment using SAR system
◮ Scatterers of interest◮ Target → Deterministic scattering◮ Tree trunks (interferences) → Deterministic scattering
◮ Others scatterers◮ Branches, foliage → Random scattering
3/24 IGARSS 2011 July 2011
SAR Imagery AlgorithmsSimulated data
Real dataConclusion and Future Work
FoPen Detection
Classical SAR
No prior-knowledge on the scatterers → isotropic and white point scatterer model
Simulated data in VV of a box in a forest of trunksReal data in VV of a truck and a trihedral in the Nezerforest
Results
◮ Low response of the target → Target not detected◮ High response of the forest → Many false alarms
4/24 IGARSS 2011 July 2011
SAR Imagery AlgorithmsSimulated data
Real dataConclusion and Future Work
FoPen Detection
Classical SAR
No prior-knowledge on the scatterers → isotropic and white point scatterer model
Simulated data in VV of a box in a forest of trunksReal data in VV of a truck and a trihedral in the Nezerforest
Results
◮ Low response of the target → Target not detected◮ High response of the forest → Many false alarms
4/24 IGARSS 2011 July 2011
SAR Imagery AlgorithmsSimulated data
Real dataConclusion and Future Work
FoPen Detection
Classical SAR
No prior-knowledge on the scatterers → isotropic and white point scatterer model
Simulated data in VV of a box in a forest of trunksReal data in VV of a truck and a trihedral in the Nezerforest
Results
◮ Low response of the target → Target not detected◮ High response of the forest → Many false alarms
4/24 IGARSS 2011 July 2011
SAR Imagery AlgorithmsSimulated data
Real dataConclusion and Future Work
New SAR processors
Approach
◮ To reconsider the SAR image generation by including prior-knowledge on thescatterers of interest
◮ To generate one single SAR image
→ Use of subspace methods
Awareness of the scattering and polarimetric properties:
1. Of the target → To increase its detection
2. Of the interferences → To reduce false alarms→Only possible if the target and the interferences scattering have different properties
5/24 IGARSS 2011 July 2011
SAR Imagery AlgorithmsSimulated data
Real dataConclusion and Future Work
Outline
SAR Imagery Algorithms
FoPen Simulated data
Real data
Conclusion and Future Work
6/24 IGARSS 2011 July 2011
SAR Imagery AlgorithmsSimulated data
Real dataConclusion and Future Work
SAR AlgorithmsCSARSSDSAROBSAROSISDSAR
Outline
SAR Imagery AlgorithmsSAR AlgorithmsClassical SAR (CSAR)SSDSAROBSAROSISDSAR
FoPen Simulated data
Real data
Conclusion and Future Work
7/24 IGARSS 2011 July 2011
SAR Imagery AlgorithmsSimulated data
Real dataConclusion and Future Work
SAR AlgorithmsCSARSSDSAROBSAROSISDSAR
SAR data configuration
SAR signal
Single Polarization p
SAR signal zp ∈ CNK
zp=
.
.
.
Double Polarization
SAR signal z ∈ C2NK
z =
.
.
.
.
.
.
8/24 IGARSS 2011 July 2011
SAR Imagery AlgorithmsSimulated data
Real dataConclusion and Future Work
SAR AlgorithmsCSARSSDSAROBSAROSISDSAR
SAR data configuration
◮ K time samples
SAR signal
Single Polarization p
SAR signal zp ∈ CNK
zp=
zp1...
Double Polarization
SAR signal z ∈ C2NK
z =
.
.
.
.
.
.
8/24 IGARSS 2011 July 2011
SAR Imagery AlgorithmsSimulated data
Real dataConclusion and Future Work
SAR AlgorithmsCSARSSDSAROBSAROSISDSAR
SAR data configuration
◮ K time samples◮ N antenna positions ui
SAR signal
Single Polarization p
SAR signal zp ∈ CNK
zp=
zp1...
zpN
Double Polarization
SAR signal z ∈ C2NK
z =
.
.
.
.
.
.
8/24 IGARSS 2011 July 2011
SAR Imagery AlgorithmsSimulated data
Real dataConclusion and Future Work
SAR AlgorithmsCSARSSDSAROBSAROSISDSAR
SAR data configuration
◮ K time samples◮ N antenna positions ui
◮ Polarization: single VV (or HH) or
SAR signal
Single Polarization p
SAR signal zp ∈ CNK
zp=
zp1...
zpN
Double Polarization
SAR signal z ∈ C2NK
z =
zHH1...
zHHN
.
.
.
8/24 IGARSS 2011 July 2011
SAR Imagery AlgorithmsSimulated data
Real dataConclusion and Future Work
SAR AlgorithmsCSARSSDSAROBSAROSISDSAR
SAR data configuration
◮ K time samples◮ N antenna positions ui
◮ Polarization: single VV (or HH) or double (HH and VV)
SAR signal
Single Polarization p
SAR signal zp ∈ CNK
zp=
zp1...
zpN
Double Polarization
SAR signal z ∈ C2NK
z =
zHH1...
zHHN
zVV1...
zVVN
8/24 IGARSS 2011 July 2011
SAR Imagery AlgorithmsSimulated data
Real dataConclusion and Future Work
SAR AlgorithmsCSARSSDSAROBSAROSISDSAR
Image generation principle
For each pixel (x , y)
Computation of the SAR response of the model
Classical model◮ White isotropic point scatterer response
Subspace models◮ Canonical element responses for all its orientations◮ Generation of the subspace
9/24 IGARSS 2011 July 2011
SAR Imagery AlgorithmsSimulated data
Real dataConclusion and Future Work
SAR AlgorithmsCSARSSDSAROBSAROSISDSAR
Image generation principle
For each pixel (x , y)
Computation of the SAR response of the model
Classical model◮ White isotropic point scatterer response
Subspace models◮ Canonical element responses for all its orientations◮ Generation of the subspace
Computation of the complex amplitude coefficient (or the coordinate vector)
◮ Least square estimation
9/24 IGARSS 2011 July 2011
SAR Imagery AlgorithmsSimulated data
Real dataConclusion and Future Work
SAR AlgorithmsCSARSSDSAROBSAROSISDSAR
Image generation principle
For each pixel (x , y)
Computation of the SAR response of the model
Classical model◮ White isotropic point scatterer response
Subspace models◮ Canonical element responses for all its orientations◮ Generation of the subspace
Computation of the complex amplitude coefficient (or the coordinate vector)
◮ Least square estimation
Intensity
◮ Square norm of the complex amplitude
9/24 IGARSS 2011 July 2011
SAR Imagery AlgorithmsSimulated data
Real dataConclusion and Future Work
SAR AlgorithmsCSARSSDSAROBSAROSISDSAR
CSAR (Classical SAR)
Modeling
No prior knowledge on scatterers of interest.White Isotropic point model rxy
SAR signal modeling
z = axy rxy + n
axy unknown complex amplitude, n complex white Gaussian noise of variance σ2
Double polarization: 2 possible models◮ trihedral scattering: rxy = r+xy
◮ dihedral scattering: rxy = r−xy
CSAR image intensity
I±C (x , y) =‖r±†
xy z‖2
σ2
Equivalence with images generated withclassical SAR processors (TDCA,Backprojection, RMA)
10/24 IGARSS 2011 July 2011
SAR Imagery AlgorithmsSimulated data
Real dataConclusion and Future Work
SAR AlgorithmsCSARSSDSAROBSAROSISDSAR
SSDSAR (Signal Subspace Detector SAR)
Target modeling
Prior-knowledge: Target is made of a Set of Plates.Target model: Low Rank Subspace 〈Hxy 〉 generated from PC plates.
x=x’
y’
z’
αy
z
y
z
x
O O
z’
x’
β
x"
z"
y"=y’
α
α
β
β
(c)(b)(a)
Signal SAR modeling
z = Hxy λxy + n
Hxy : orthonormal basis of 〈Hxy 〉, λxyunknown amplitude coordinate vector.Double polarization:2 possible target subspaces
◮ trihedral scattering: Hxy = H+xy
◮ dihedral scattering: Hxy = H−xy
11/24 IGARSS 2011 July 2011
SAR Imagery AlgorithmsSimulated data
Real dataConclusion and Future Work
SAR AlgorithmsCSARSSDSAROBSAROSISDSAR
SSDSAR (Signal Subspace Detector SAR)
R. Durand, G. Ginolhac, L. Thirion-Lefevre, and P. Forster, “New SAR processor based on matched subspace
detectors,” IEEE TAES, Jan 2009.
F. Brigui, L. Thirion-Lefevre, G. Ginolhac and P. Forster, “New polarimetric signal subspace detectors for SAR
processors,” CR Phys, Jan 2010.
Goal: Improvment of target detection.
SSDSAR image intensity
IS(x , y) =‖H†
xy z‖2
σ2
PHxy = Hxy H†xy : orthogonal projector into 〈Hxy 〉.
< H >
< J >
P zH
z
11/24 IGARSS 2011 July 2011
SAR Imagery AlgorithmsSimulated data
Real dataConclusion and Future Work
SAR AlgorithmsCSARSSDSAROBSAROSISDSAR
OBSAR (Oblique SAR)
Interference modeling (Trunks)
Prior-knowledge: Trunks are dielectric cylinders lying over the ground.Interference model: Low Rank Subspace 〈Jxy 〉 generated from dielectric cylinders lyingover the ground.
x’
y’
z’=z
y
z
x
O
δ
δ
δ γ
γ
γ
x"
z"
y"=y’O O
(a) (b) (c)
Signal SAR modeling
z = Hxy λxy + Jxy µxy + n
Jxy : orthonormal basis of 〈Jxy 〉, µxy unknown amplitude coordinate vector.Double polarization: 1 possible interference subspace
12/24 IGARSS 2011 July 2011
SAR Imagery AlgorithmsSimulated data
Real dataConclusion and Future Work
SAR AlgorithmsCSARSSDSAROBSAROSISDSAR
OBSAR (Oblique SAR)
F. Brigui, G. Ginolhac, L. Thirion-Lefevre, and P. Forster, “New SAR Algorithm based on Oblique Projection for
Interference Reduction,” IEEE TAES, submitted.
Goals:◮ Increase of target detection.◮ Reduce false alarms due to deterministic interferences.
OBSAR image intensity
IOB(x , y) =‖H†
xy EHxy Jxy z‖2
σ2
EHxy Jxy = PHxy (I − Jxy (J†xy P⊥Hxy
Jxy )−1J†xy P⊥Hxy
):
oblique projector into 〈Hxy 〉 along the directiondescribed by 〈Jxy 〉.