multi-chromatic analysis of insar data: validation … · 2017. 11. 29. · fringe – esa esrin,...
TRANSCRIPT
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 1
MULTI-CHROMATIC ANALYSIS OF INSAR DATA: VALIDATION AND POTENTIAL
Fabio Bovenga, Vito Martino Giacovazzo, Alberto Refice, Nicola Veneziani, Raffaele Vitulli (ESA ESTEC)
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 2
Contents
• Multi Chromatic Analysis (MCA) background
• Implementation aspects.
• First validation on airborne and spaceborne SAR sensors.
• Comments and conclusions.
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 3
InSAR principles
MR
tM
ϑ
⊥B
||B
Swath
Gruond range
tS
SR
sB
H
Topography Atmosphere Deformation
cdefAtmc frxRrxRRyxHB
cfrxR
crx ⋅⎥
⎦
⎤⎢⎣
⎡∆+∆+−=⋅∆⋅−=Φ ⊥ ),(),(
sin),(4),(4),(
0 θππ
SM RRR −=∆
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 4
InSAR principles
),(2),(),( rxkrxrx W ⋅+Φ=Φ π
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 5
Original idea
• Madsen and Zebker:
– φabs is used to compute the K2π necessary to recover the absolute phase
– Averaging is required to keep the noise low and attenuate effect of factor f0 / (f+-f-)
( ) diffabs fff φφ
−+ −= 0
( ) ddiff tff ⋅−⋅=−= −+−+ πφφφ 2
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 6
Multi-Chromatic Approach (MCA) principles
• MCA of InSAR processing involves the computation of a set of several equal-resolution interferograms from SAR sub-look images of bandwidth Bp centred at different carrier frequencies fi :
• Only wrapped phase values are known:
iic fRcfR
c⋅∆−=Φ→⋅∆−=Φ ππ 44
]2/ , 2/[ iiiiici BfBfdfff +−∈+=
),(),(2),(),(2),(),( rxrxkrxrxkrxrx iWi
Wii Φ+⋅−=Φ→⋅+Φ=Φ ππ
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 7
Multi-Chromatic Approach (MCA) principles
iiWi frxCrxCfrxRc
rxkrx ⋅+=⋅∆⋅−⋅−=Φ ),(),(),(4),(2),( 10ππ
Height mesurement
),(4
),( 1 rxCcrxR ⋅−=∆π
Absolute phase measurement /Finges classification
π2),(),( 0 rxCrxk −=
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 8
Multi-Chromatic Approach (MCA) principles
• The basic MCA algorithm, for the height measurement, consists ofperforming standard InSAR processing on sub-band signals, followed by a point-wise estimation of C0 and C1 through a linear regression method.
• The height of any pixel is retrieved independently from the others and no phase unwrapping step is needed in the spatial domain. The resolution of the resulting interferogram depends on the spectral sub-band width, Bp.
• The inter-band coherence or the multi-frequency phase STD for the MCA and can be assumed as a quality index of the measurements:
• Reliable measurement can be performed on the targets showing a good linear phase trend vs. frequency (PSfd):
( ) ( ) ( )[ ]∑=
⋅−−Φ=Φ
fN
iii
f
frxcrxcrxN 1
210 ,,,
1 σ
( ){ }thfd yxyxP ΦΦ ≤= σσ ),(|,
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 9
MCA on 100 MHz AES dataset (Langnau)
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 10
MCA potentials
• Height retrieval (C1) / Finge classification (C0)
• Support / enforce standard PU algorithms
• Absolute ranging from single SLC
• Coherent scatterers identification
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 11
MCA implementation
• First tested in simulation and applied on ERS data: satellite sensor available didn’t allow a reliable application.
• However, the technique appears optimally suited for the new generation of satellite sensors, which operate with larger bandwidths (TerraSAR-X or COSMO-SkyMed) → new impulse to the experimentation!
• First MCA implementation worked on RAW data and generate sub-bands images by focusing portions of the available bandiwidth.
• But data policies of these missions do not foresee the delivery of RAW data thus imposing to perform the MCA starting from SLC images.
• MCA form SLC data has been investigate both theoretically and byprocessing real data.
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 12
MCA processing
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 13
MCA processing: coregistration
• ISSIA MCA processing chain is designed to have in input a couple of SLC images already co-registered to avoid increasing computational efforts related to the coregistration between Master and Slave images for each sub-bands.
• Co-registration step can be simply modeled as e shift in the space wichintroduce a phase ramp in the spectrum.
• When the sub-bands filtering is applied to the Slave image after interpolation, an additional phase term proportional to the frequency arises in the sub-band images:
( ) ( ) ( ) )()()()()()()( 21211 ttIIFTeFTfHIFTeFTfHIFTFTtI iitfjitfjiCCi ∆−=⋅=⋅⋅=⋅= ∆−−∆−−− ππ
( ) tfjtfji
Ci
ii eettItI ∆⋅−−⋅∝∆−∠=∠ πτπ 22)()(
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 14
MCA processing: coregistration
• Then, the interferometric phase results:
• In order to infer the proper path difference it is needed to take into account the phase term related to the coregisration shift. This operation is trivial and not computationally expensive:
• The range shift matrix shr(p) is already available as side product of the co-registration step.
( ) irsirsfpsh
fj
iSiM
fpshf
j
ii eIIepINTpINT⋅
∗⋅
⋅⋅=⋅→)(2
,,
)(2
)()(ππ
( ) irs
ii fpshfpR
cfptpp ⋅⎟⎟
⎠
⎞⎜⎜⎝
⎛+∆−=⋅∆+∆⋅−=Φ )(2)(4)()(2)( ππτπ
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 15
MCA processing: de-Hamming
-6 -4 -2 0 2 4 6
x 107
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
Frequency [Hz]
FT a
mpl
itude
MASTER TSX-2008090: Spectrum de-hamming
OriginalHammingde-HammingFiltered
• In order to avoid asymmetry in the spectrum, before the pass-band filtering a de-Hamming filter is applied to the SLC data which consists of weighting the range spectrum of the data by the function:
H f( )= 1α − 1−α( )cos 2πf( )
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 16
MCA processing: height retrieval validation
• The InSAR phase can be expressed in terms of a reference phase (Φref) which accounts for both the difference between the slant range geometry and a reference ellipsoid and, the phase related with the elevation wrt this ellipsoid (ΦH):
• In order to have available a topographic information in SAR geometry, the reference DEM has back-projected in slant range through standard InSARprocessing chain: HDEM(x,r)
⎥⎦⎤
⎢⎣⎡ +Φ
⋅⋅−≈Φ−Φ+Φ=⋅=Φ rHpHf
cpppfCp i
shrefHii δ
π2
)(4)()()()( 1
)(2)()(4
)( 1 pHprpCcpH Φ⋅⎥⎦
⎤⎢⎣⎡ +⋅−= δπ
)(2
)(4
pshfcp
fcr r
s
ref
c
−Φ−=π
δ
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 17
MCA processing: TSX dataset
• A first validation started by using a TerraSAR-X dataset acquired on Langnau. This choice was driven by the possibility to obtain the data needed for validation by using a standard InSAR pre-processing.
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 18
MCA processing: TSX dataset
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 19
MCA Pre-processing: TSX dataset
010
2030
4050
60 0
50
100
150
700
750
800
850
900
Range
Azimuth
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 20
MCA processing: TSX dataset
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 21
MCA processing: TSX dataset
0 2 4 6 8 10 12
x 107
0
5000
10000
15000Slave-SLC: Pass-band spectrum, fci =-25 MHz
Frequency
Am
plitu
de
Slave-SLC: FFT along range, fci =-25 MHz
Range FFT samples
Azi
mut
h
100 200 300 400 500 600 700 800 900
50
100
150
200
250
300
350
400
450
500
Interferometric Phase - Sub-band:1
Range
Azi
mut
h
50 100 150 200 250 300 350 400 450 500
100
200
300
400
500
600
700
800
900
0 2 4 6 8 10 12
x 107
0
5000
10000
15000Slave-SLC: Pass-band spectrum, fci =0 MHz
Frequency
Am
plitu
deSlave-SLC: FFT along range, fci =0 MHz
Range FFT samples
Azi
mut
h
100 200 300 400 500 600 700 800 900
50
100
150
200
250
300
350
400
450
500
Interferometric Phase - Sub-band:2
Range
Azi
mut
h
50 100 150 200 250 300 350 400 450 500
100
200
300
400
500
600
700
800
900
0 2 4 6 8 10 12
x 107
0
5000
10000
15000Slave-SLC: Pass-band spectrum, fci =25 MHz
Frequency
Am
plitu
de
Slave-SLC: FFT along range, fci =25 MHz
Range FFT samples
Azi
mut
h
100 200 300 400 500 600 700 800 900
50
100
150
200
250
300
350
400
450
500
Interferometric Phase - Sub-band:3
Range
Azi
mut
h
50 100 150 200 250 300 350 400 450 500
100
200
300
400
500
600
700
800
900
0 500 1000 1500 2000 25000
0.2
0.4
0.6
0.8
1
1.2
1.4Pass-band Filter Responce
0 500 1000 1500 2000 25000
0.2
0.4
0.6
0.8
1
1.2
1.4Pass-band Filter Responce
0 500 1000 1500 2000 25000
0.2
0.4
0.6
0.8
1
1.2
1.4Pass-band Filter Responce
Kernel Spectrum around B/2 Image Spectrum around 0 Φi(x,r)
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 22
MCA processing: TSX dataset
( ) ( ) π
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 23
MCA processing: TSX results
σφth=0.02 rad
• A useful analysis for validate the output of the MCA processing is the inspection of the distribution of the pixels in PSfdsuperimposed on the SAR amplitude as well as on an optical image.
• Targets in PSfd drop mainly on pixels which show high amplitude values or, in the optical image, on resolution cell which correspond to man made objects (as buildings).
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 24
MCA processing: TSX results
• Trends before shift compensation for pixel with σφ=0.0038 rad
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 25
MCA processing: TSX results
=-113.0975 m
• The trends show the effect of shift compensation for pixel withσφ=0.0038 rad
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 26
MCA processing: TSX results
• Pixels with σφ
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 27
MCA processing: TSX results
• In order to validate the results, we use external DEM back projected onto SAR Master geometry: HDEM(x,r).
• The difference between these height values is due to a deviation of the interferometric phase Φ wrt the expected value :Φ̂
),(),(),(),(ˆ),(),( rxrxrxrxrxrx noiseprocAPSR Φ+Φ+Φ=Φ−Φ=Φ
-2500 -2000 -1500 -1000 -500 0 500 1000 1500 2000 25000
0.5
1
1.5
2
x 104 Expected-estimated absolute phases (pick value = -30 rad) • In the present case it is not possible to
discriminate between the different contributions to the residual phase.
• Temporal decorrelation (Bt = 33 d) →increase the noise level.
• Additional path induced by the atmosphere → bias in the estimation of absolute height (a variation of λ/2 causes an error of 50 m in H ).
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 28
MCA processing: TSX results
-2500 -2000 -1500 -1000 -500 0 500 1000 1500 20000
2
4
6
8
10
12
14
16
18Height error distribution at th < 0.02 (pick value = -79 m)
-2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5
x 104
0
0.5
1
1.5
2
x 104Expected-estimated heigths - pick value = -229 m (no th, no APSc)
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 29
∆RAPS
50 100 150 200 250 300 350 400 450 500
100
200
300
400
500
600
700
800
900
1000 0.061
0.062
0.063
0.064
0.065
0.066
0.067
0.068
0.069
|∆RDEM|
50 100 150 200 250 300 350 400 450 500
100
200
300
400
500
600
700
800
900
10000.24
0.245
0.25
0.255
0.26
0.265
0.27
0.275
0.28
0.285
0.29
MCA processing: TSX results
δr (m)
50 100 150 200 250 300 350 400 450 500
100
200
300
400
500
600
700
800
900
1000 0.305
0.31
0.315
0.32
0.325
0.33
0.335
0.34
0.345
0.35
( )( )⎥⎦
⎤⎢⎣⎡ ∆−∆−∆
⋅−⋅≈
≈Φ−Φ−Φ+⋅=Φ∧
),(),(),(4),(
),(ˆ),(),(),(),(
1
1
rxRrxRrxRc
ffrxC
rxrxrxfrxCrx
refshHc
c
Hrefshc
R
π
• Possible rough atmospheric signal estimation by using standard pre-processing side products and SRTM DEM (90x90 m)
The spectral density shows an average trend close to f(-5/3) which is typical of signals related to the wet component of the troposphere.
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 30
MCA processing: TSX results
• Performing the estimation on very coherent targets by means of asevere threshold on σΦ (
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 31
MCA processing: TSX results
• The true problem remains the difficulty to separate different sources of phase signals without the help of ancillary data or further temporal analysis.
• Also the bandwidth value represents a limiting factor in the MCAperformace.
• In order to evaluate the impact of this parameter on height retrieval through MCA processing, the ariborne airborne AES-1 dataset has been used: it is not affected by neither atmosphericdistortion nor temporal and geometrical decorrelation and, provides the possibility to explore different values of bandwidth from 100 to 400 MHz
• The experiment showed that a bandwidth of least 300 MHz seems to be required in order to obtaine reliable results height measurements.
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 32
MCA processing: AES-1 results
• Dataset acquired by airborne AES-1 sensor appears very promising for MCA testing
• A multi-channel In-SAR sensor operating at X-band, with a total radar bandwidth of 400 MHz resulting from four non-overlapping channels, each 100 MHz wide, whose central frequencies are in GHz {9.4 9.5, 9.6, 9.7}.
• For this dataset the normal baseline B⊥=0.75 m thus ensuring a very limited geometrical decorrelation effect.
• The two antenna acquire simultaneously:– NO temporal decorrelation– NO atmospheric contribution to the interferometric phase are avoided
),(),(),(),(),(0
rxrxrxrxrx procnoiseprocAPSR Φ≈Φ+Φ+Φ=Φ→φσ
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 33
MCA processing: AES-1 results
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 34
MCA processing: AES-1 results
• However, for airborne dataset is not possible to perform InSARpre-processing by using standard InSAR chain designed for satellite missions → Φref(x,r), Φ2H(x,r) , HDEM(x,r) are not determined properly.
• Moreover, the analysis of the data pointed out the presence of adelay introduced by the hardware between M and S. Since, the analysis of the ancillary data did not allow to define the actual value of this delay, it result impossible to perform absolute interferometric pahse estimation.
• For these reasons the experiment with AES data was performed by analysing the data in slant range geometry without a direct comparison wrt the SRTM height profile.
• Anyway, this limitation didn’t prevent to verify how the MCA performances depend on the available bandwidth.
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 35
MCA processing: AES-1 results
11 fringes
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 36
MCA processing: AES-1 results11 K values K(x,r)
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 37
MCA processing: AES-1 resultsK(x,r)
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 38
MCA processing: parametric analysis
• Relevant parameters: B , Nf , Bp , Be• Estimation of the expected noise as function of the processing parameters.
• In the hypothesis of ideal metallic planar reflector:
( )
( )Φ
==
=
⎟⎟⎠
⎞⎜⎜⎝
⎛−
=
∑∑
∑σσ 2
11
2
1
2
0ff
f
N
ii
N
iif
N
ii
C
ffN
f
( )Φ
= =∑ ∑ ⎟⎟
⎠
⎞⎜⎜⎝
⎛−
= σσ f fN
i
N
iiif
fC
ffN
N
1
2
1
21
clutterA
ASCR
0
22
4
σλπ⋅
=
)m (0.5x0.5 reflector planarmetallic the of areaAcell resolution the of areaA
1)( section cross radar normalisedσwhere
2clutter
0
=
=
==
SCR⋅=Φ 2
1σ
),(4
),( 1 rxCcrxR ⋅−=∆π
π2),(),( 0 rxCrxk −=
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 39
MCA processing: parametric analysis
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 40
MCA processing: parametric analysis
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 41
MCA processing: parametric analysis
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 42
MCA processing: comments
• It can be noted that the K values map obtained by processing a bandwidth of 400 MHz reproduces properly the fringe pattern.
• At least a 300 MHz bandwidth seems to be required to provide reliable results.
• For B=100 MHz the correlation with the fringe pattern disappears. This value appears inadequate for reliable height inference.
• Therefore, this experiment confirms that, in the case of TSX dataset, where the phase information is corrupted by temporal and geometrical decorrelation and by the presence of atmospheric signal, a bandwidth of 100 MHz leads to unreliable height.
• The results are confirmed by the parametrical analysis obtained through theoretical modeling.
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 43
MCA processing: work in progress
• Experiment with TSX 300 MHz bandwidth Spotlight acquisition.
• Experiment with COSMO-SkyMED 130 MHz bandwidth on desert area.
• More reliable figure for frequency coherent target (σΦ is not robust)
• Multi-temporal dataset exploration for estimate the atmospheric influence.
-
FRINGE – ESA ESRIN, 30/11/2009 CNR-ISSIA Bari 44
The End
Work supported by ESA ESTEC Contr. N. 21319/07/NL/HE.
TerraSAR-X images are provided by DLR under a Scientific Use License for the proposal MTH0397.
Sample AES-1 data is courtesy of SARMAP
Thank you