Fabrizio LombardiniFederico Viviani

DIFF-TOMO OPENING OF THE URBAN SAR PIXEL: SINGLE-LOOK 4D AND NON-UNIFORM MOTION

“5D”

EXTENSIONS

Part of this work has been supported by ASI project n. I/065/09/0

University of Pisa

Recall

of

Tomographic

SAR techniques-

3D SAR Tomography-

4D Differential

Tomography

(3D +time)

3D /4D/”5D”

Tomographic

advances

for

urban

areas:

Full horizontal

resolution

Tomography-

Single-look

adaptive

Tomography, results

with

ERS data-

Tests

of

single-look model-based

Tomography

Full resolution

processing extensions

of

Differential

Tomography-

Adaptive

processing, first model-based

processing test-

First test of

adaptive

Diff-Tomo

for

non-uniform

motions

(5D)

First Tomographic

trial with

C-S data; Differential

Tomography

extensions

for

forest

applications

Conclusions

and future work

Outline

2

3D SAR TomographyFrom 2.5D imaging (InSAR) to full 3D imaging…Multibaseline (MB) technique for imaging (z profiling) of elevation-distributed scatterers (multiple “garbled” scatterers, semitransparent volume scattering layers)

MB array response to backscattering from height h

Spatial frequency related to the scatterer normal height s

Height profiling by elevation beam-steeringHowever…The limited

and irregular

baseline

distribution

causes

anomalous

height

sidelobe behaviour

and limited

resolution

Many solutions have been proposed to counteract these problems…

• Array interpolation (gap filling)+ Fourier processing (IBF) [DLR ‘00]

• Regularized inversion approach [CNR-IREA ‘02]

• Adaptive superresolution tomography, model-based superresolution

• Volumetric coherence-based inversion approach [Cloude ’07]

• Sector Interpolation [UniPi ’07]

• Compressive sensing [Univ. Naples ‘09]

Ground range

Height, hNormal height, s

Slant range

12

KBK

3

a

[Lombardini, TGARS’05]

Define a velocity-dependent temporal frequency: (D-InSAR)

2-D Fourier relationSpatial-temporal spectral estimation

Multitemporal multibaseline cmplx data

Cmplx amplitudeelevation-velocity

distribution

Sparse 2D: multi-ant. air./sat. cluster

& multi-pass (multistatic)

“1D” Curvilinear: plain multi-baseline by

multi-pass, airborne/sat. (monostatic)

2D support in 2D baseline-time plane deeply

exploited

2D elev.-vel. sidelobe challenge; handling by advanced spectral estimators

D-InSAR concept Tomo-SAR concept Conv. acquisition New processing

DiffDiff--TomoTomo**, it , it ““opensopens”” the SAR pixelthe SAR pixel extractingjoint height and dynamical information of superimposed moving

scatterers(“4D imaging”, 3D+time)

New output!

Define an elevation-dependent spatial frequency: (Tomo-SAR)

Differential SAR TomographyFrom 3.5D imaging (D-InSAR) to full 4D imaging…

* patent pending4

Advances of 3D Adaptive Tomography Adaptive Tomo: good superresolution/sidelobe reduction capabilities,

yet it requires coherent multilookingMoving from multilook to single-look processing…Full horizontal resolution is more desirable in urban remote sensing applications

SectorInterpolation

y

SOI

1 lookPreconditioner

Spatial filter

can operate at full horizontal resolution deterministic version for uniform data

Tomo image

Sidelobe reduction and height superresolutionat full range-azimuth resolution

No increased computational burden!

… and to single-look adaptive Differential Tomographic processing…the proposed solution can be extended to the framework of Differential Tomography 5

Real data results Single - look processing

Tomographic profiles from adaptive and model-based tomography

superresolutionmain lobe width gain=43.5

• ERS 1/2 dataset • K=43 tracks• total baseline 1500 m• time span 5 years• Rayleigh height res. 6.4 m

Rome, Cinecittà

Model based Tomography

Adaptive Tomography

Clipped amplitudes

6

( Fourier )

Superresolution separation gain up to 4

Real data results

Single - look processing

Superresolution capabilities of model-based tomography

Tomographic profiles from double scatterer cells(separation –wise ordered and aligned)

Rayleigh resolution limit (6.4m)

Δh=1.7 mΔh=10 m

• Identification of double scatterers with separation 3.7 times below the Rayleigh resolution limit

7

Single-look 4D Adaptive Differential Tomography

The new method extended to the framework of Diff-Tomo

Spatial-temporalInterpolation

PreconditionerSpatial-temporal filter

Diff-Tomo image

1 look

y

SOI

Cinecittà dataset : Height – Velocity PSF

Adaptive Diff-TomoFourier Processing

PSF (PSF)Amplitudes 8

Adaptive Diff-Tomo

Double scatterer cell

Adaptive Diff-Tomo

Single scatterer cell

Single-look Adaptive Differential Tomography

Double scatterer cell Double scatterer cell

Adaptive Diff-Tomo Adaptive Diff-Tomo

Elevation-velocity power distribution (3-th and 4-th dimensions)

Cinecittàdataset

First results …

9

• Identification of double scatterers below the Rayleigh resolution limit

Double scatterer cell

Model-Based Diff-Tomo

Single scatterer cell

Single-look Adaptive Differential Tomography

Model-Based Diff-Tomo

Elevation-velocity power distribution (3-th and 4-th dimensions)

Cinecittàdataset

First results …

9

“5D”

Diff-Tomo for non-uniform motionsExtension of adaptive Diff-Tomo

to extract different (average) accelerations of multiple scatterers Semi-parametric adaptive Diff-TomoExploits a model for non-uniform motions

Simulated analysis Cinecittà baseline pattern 1 steady scatterer 1 moving scatterer with linear uniform motion

- Initial velocity=2.8 mm/year- Velocity increment=5.6 mm/year

Semi-parametric Diff-Tomo output(after optimization over accelearation parameter)

Estimated (average) acceleration (5-th dimension)

“5D” focusing achieved!

the output is in the joint range-azimuth-elevation-

velocity-acceleration domain

Recovery of power loss due to non-uniform motion

Velocity Increment (m

m/year)

10

“5D”

Diff-Tomo for non-uniform motionsExtension of adaptive Diff-Tomo

to extract different (average) accelerations of multiple scatterers Semi-parametric adaptive Diff-TomoExploits a model for non-uniform motions

Simulated analysis Cinecittà baseline pattern 1 steady scatterer 1 moving scatterer with linear uniform motion

- Initial velocity=2.8 mm/year- Velocity increment=5.6 mm/year

Semi-parametric Diff-Tomo output(after optimization over accelearation parameter)

Estimated (average) acceleration (5-th dimension)

Velocity Increment (m

m/year)

10

“5D” focusing achieved!

the output is in the joint range-azimuth-elevation-

velocity-acceleration domain

Recovery of power loss due to non-uniform motion

“5D”

Diff-Tomo for non-uniform motionsExtension of adaptive Diff-Tomo

to extract different (average) accelerations of multiple scatterers Semi-parametric adaptive Diff-TomoExploits a model for non-uniform motions

Simulated analysis Cinecittà baseline pattern 1 steady scatterer 1 moving scatterer with linear uniform motion

- Initial velocity=2.8 mm/year- Velocity increment=5.6 mm/year

Estimated (average) acceleration (5-th dimension)

Velocity Increment (m

m/year)

Statistics

Steady Scatterer Moving Scatterer

Bias Initial Velocity (mm/yr)

0.2 0.1

RMSE Initial Velocity (mm/yr)

0.2 0.7

Bias Velocity Variation (mm/yr)

-1 -0.4

RMSE Velocity Variation (mm/yr)

0.2 0.9

10

Δv=5.4 mm/year

Δv=0.2 mm/year

Real data results

5D Diff-Tomo

1° scatterer

2° scatterer

Velocity Increment (m

m/year)

Cinecittàdataset

11

Double Scatterer cell

4D Diff-Tomo

11

Real data results

5D Diff-Tomo

1° scatterer

2° scatterer

∆v=-3 mm/yr

∆v= -3.3 mm/yr

Velocity Increment (m

m/year)

Cinecittàdataset

Double Scatterer cell

4D Diff-Tomo

First Tomographic trial with COSMO-SkyMed data

12

• COSMO – SkyMed dataset• K=26 tracks• total baseline 1000 m• time span 1 year

Naples

13

The Diff-Tomo framework can also be applied to volumetric scenarios:Novel functionalities for probing 3D forest structure and internal dynamics

Tomography robust to temporal decorrelation

Subcanopy ground subsidence monitoring

Coherence separation

0 0.2 0.4 0.6 0.8 1

0

10

20

30

40

50

Hei

ght [

m]

Normalized

Tomo MUSICDiff-Tomo Gen. MUSIC

Sensible mitigation of blurring effects from temporal

decorrelation

Decoupling of forest volume from ground scatterer:

subcanopy subsidence may be revealed

Different temporal decorrelation mechanisms can be identfied in the

same resolution cell

Robust tomographic profiles

Estimated ground subsidence

BioSAR P-band data set

Diff-Tomo :Temporal decorrelation signature

Height-velocity-temporal bandwidth“5D” functional

[Lombardini-Cai, PolInSAR ‘11]

[Lombardini-Cai, IGARSS ‘08]

13

The Diff-Tomo framework can also be applied to volumetric scenarios:Novel functionalities for probing 3D forest structure and internal dynamics

Tomography robust to temporal decorrelation

Subcanopy ground subsidence monitoring

Coherence separation

0 0.2 0.4 0.6 0.8 1

0

10

20

30

40

50

Hei

ght [

m]

Normalized

Tomo MUSICDiff-Tomo Gen. MUSIC

Sensible mitigation of blurring effects from temporal

decorrelation

Decoupling of forest volume from ground scatterer:

subcanopy subsidence may be revealed

Different temporal decorrelation mechanisms can be identfied in the

same resolution cell

Robust tomographic profiles

Estimated ground subsidence

BioSAR P-band data set

Diff-Tomo :Temporal decorrelation signature

Height-velocity-temporal bandwidth“5D” functional

See the poster this Tuesday afternoon!

[Lombardini-Cai, PolInSAR ‘11]

[Lombardini-Cai, IGARSS ‘08]

Conclusions

• Tomographic/Differential-Tomographic techniques are emerging methodologies for description and monitoring of garbled scattering urban areas

• 3D Multilook Tomography methods have been extended to the single look case

• Height superresolution

at full horizontal resolution has been achieved mantaining the low computational burden of adaptive and model-based Tomography

• This new single look processing has been extended to the framework of 4D Differential Tomography, both adaptive and model-based

• Diff-Tomo accounting for non-uniform motions: first encouraging results achieved of 5D adaptive Differential Tomographic imaging

Future works• Development of these Tomographic techniques for application to the COSMO-SkyMed data

• Expanding 4D/5D processing tests, and comparisons with other single look techniques

• Tuning of the Differential Tomography functionalities for forest applications14

Thanks for your attention…

15

Single-look 4D Adaptive Differential Tomography

Adaptive Diff-Tomo

Double scatterer cell

Adaptive Diff-Tomo

Single scatterer cell

16

Model-based Diff-Tomo

Single scatterer cell

First result of automated identification of scatterers through single-look adaptive differential tomography and data-domain fitting

17

First result of automated identification of scatterers through single-look adaptive differential tomography and data-domain fitting

18

“5D”

Diff-Tomo -

Real data results

5D Diff-Tomo

1° scatterer

2° scatterer

∆v=0.7 mm/yr

∆v= -

0.9 mm/yr

Velocity Increment (m

m/year)

Cinecittàdataset

19

4D Diff-Tomo

Diff-Tomo for non-uniform motions

• Simulated analysis• Two scatterers at same height and initial velocity• Possibility to resolve in the acceleration domain• “FULL” 5D

20