FRINGE 2007 Workshop - European Space Agency Multidimensional SAR SAR Imaging Imaging: : Multidimensional Studies in the in the Framework Framework Studies of LIMES Project of LIMES Project M. Costantini 1 , G. Fornaro 2 , F. Lombardini 3 , M. Pardini 3 , F. Serafino 2 , F. Soldovieri 2 1 Telespazio S.p.A. Via Tiburtina, 965 - I-00156 Roma (Italy) 2 IREA-CNR Via Diocleziano,328 I-80124 Napoli (Italy) 3 Dip. di Ingegneria dell’Informazione, Università di Pisa Via G. Caruso, 16 - I-56122 Pisa (Italy)
Overview � The LIMES project � Motivation � 3D tomographic imaging, ERS results � 4D differential tomography, ERS results � Performance evaluation, satellite clusters � Conclusions
LIMES Land and Sea Integrated Monitoring for European Security Sixth Framework Program GMES Security Development of satellite-based services providing relevant information and decision-support tools in relation to: � Critical infrastructure monitoring � Organization and distribution of humanitarian relief & reconstruction � Surveillance of the EU borders (land and sea) � Surveillance and protection of maritime transport for sensitive cargo � Protection against emerging security threats Among the main users involved in the project there are Civil Protection organizations, FRONTEX, EU and International Agencies In the framework of LIMES project, SAR 3D/4D Tomography is experimented for supporting Critical Infrastructure Surveillance Critical regasification plants and pipelines in Spain considered as test areas
Motivation • SAR interferometry • accurate determination of scatterer locations • Multibaseline InSAR • SAR differential interferometry • precise tracking of their movement • Multitemporal interferogram stacking • • Persistent scatterers � Surface penetration Superposition of responses from � Steep ground topography ( layover ) multiple scatterers in the same pixel � High spatial density of strong scatterers Need for more sophisticated techniques � Multibaseline data 3D SAR Tomography Tomography 3D SAR � Separation in elevation of scattering [Pasquali-Prati-Rocca et al., IGARSS ’95] contribution within a single pixel [Reigber-Moreira, IEEE-TGARS ’00] � Full 3D Imaging 4D SAR Imaging Imaging 4D SAR � Multibaseline multitemporal data (differential differential tomography tomography) ) ( � Joint elevation-velocity reconstruction [Lombardini, IEEE-TGARS ’05]
3D Imaging: SAR Tomography flight direction g N � Data acquired at the n th antenna (pass): elevation g n g 0 n-th orthogonal S N Backscattering baseline profile S n SPATIAL COMPLEX AMPLITUDE SPECTRUM SPATIAL COMPLEX AMPLITUDE SPECTRUM [ Note: equivalent to the classical imaging concept ] S 0 After collecting all data, the problem is the x inversion of a simple semi-discrete linear operator: z s � Beamforming r The inversion can be afforded � SVD with different algorithms Need for data calibration (removal of atmospheric � Adaptive beam. (linear, regularized, adaptive, variations, scene deformations, …) � MUSIC parametric, …) � ……
3D: Real data experiments � San Paolo stadium, Naples ERS-1/2 data, 63 images Baseline span: 1700 m, height resolution 5.5 m Temporal span: ∼ 10 years Singular Value Decomposition Adaptive beamforming SVD SAR image (single look) (5 azimuth looks) (5 azimuth looks) height range azimuth azimuth
4D: Differential Tomography g NB -1 ,NT -1 � Multistatic system: data acquired at the n th g 0, NT -1 track and the m th pass: flight direction g n,m S NB -1 ,NT -1 g NB -1,0 m-th pass n-th orthogonal Elevation-velocity baseline backscattering S 0 ,NT -1 g 0,0 profile S n,m SPATIO- -TEMPORAL COMPLEX AMPLITUDE SPECTRUM TEMPORAL COMPLEX AMPLITUDE SPECTRUM SPATIO r n,m ( s ) S NB -1,0 After collecting all data, we have again: S 0,0 x s z v � 2D Beamforming � 2D SVD N B : number of tracks per pass � 2D Adaptive beamforming r N T : number of passes � ……
4D: Experimental results (1) 2D Beamforming “Mergellina”, Naples Single scattering mechanism 30 tracks Baseline span: 1066 m, height resolution 8.8 m Time span: ∼ 6 years 2D Adaptive beamforming amplitude Velocity (mm/yr) Velocity (mm/yr) Theoretical point-spread function
4D: Experimental results (2) 2D Adaptive beamforming 2D Capon 2D Capon TOMO-DOPPLER 2D Beamforming 2D Beam. TOMO-DOPPLER 60 0.2 60 50 50 0.18 1.2 San Paolo stadium, Naples 40 40 0.16 30 Double scattering mechanism 30 1 0.14 20 20 Elevation (m) Elevation (m) Elevation (m) 0.12 10 Elevation (m) 10 0.8 ∼ 10 m 0 0.1 0 2D Adaptive beam., multilook 0.6 -10 -10 0.08 (9 looks) -20 -3.1 mm/yr -20 0.06 0.4 -30 -30 0.04 -40 -40 0.2 0.02 -50 -50 0 -10 -5 0 5 10 -10 -5 0 5 10 Doppler (mm/year) Velocity (mm/yr) Doppler (mm/year) Velocity (mm/yr) � Scatterers well resolved in height � Estimation of deformation velocity consistent with independent measures � Reduced SLL w.r.t. 2D Fourier beamforming, but higher sensitivity to miscalibration residuals � No equal velocity constraints (equal velocity case: [Ferretti-Bianchi-Prati-Rocca, EURASIP JASP ’05])
4D: Experimental results (3) “Vomero”, Naples ERS-1/2, 58 passes, ∼ 10 years temporal span SVD single look - Single scatterers - - Double scatterers - � Scattering mechanisms can be separated � Automatic single/double scatterer identification also tested
4D: Experimental results (4) San Paolo Stadium, Naples SVD single look - Single scatterers - - Double scatterers -
Imaging Capabilities and Satellite Clusters (1) Typical poor and irregular High sidelobes in the 3D/4D High sidelobes in the 3D/4D baseline/time sampling reconstructed profile profile reconstructed Double speckled compact scatterers Single track per pass SNR = 15, 12 dB 3 tracks per pass 32 looks 2D Beamforming 2D Adaptive 2D Adaptive 2D Beamforming beamforming beamforming Velocity Velocity Elevation Velocity Elevation Elevation Velocity Elevation
Imaging Capabilities and Satellite Clusters (2) 2 antennas per pass Larger SV dynamic Acquisition grid (baseline/time) Singular value (SV) distribution 58 passes, orth. baseline separation 150 m
Accuracy Bounds Evaluation � Algorithm performance judgement � Characterization of precision limits Tools from from information information theory theory: : Tools � 3D Cramér-Rao Lower Bound (CRLB) Given a statistical model for the [Gini-Lombardini-Montanari, IEEE-Tr. on AES ’02] data vector g , bounds can be evaluated for the 4D estimation of � 3D Hybrid CRLB (HCRLB) scatterer scatterer elevations elevations and line of line of takes into account possible miscalibration residuals [Pardini-Lombardini-Gini, IEEE-TSP, accepted for publication] sight velocities velocities sight � ERS-1/2, 58 passes � 10 looks � Double speckled scatterers, large critical baseline ( b ⊥ TOT / b C = 0.05, classical triangular-shaped spatial decorrelation) � Possible temporal decorrelation, τ c = 2 months (exponential decorrelation model) [Lombardini-Griffiths, IEE Meeting on RS Sign. Proc. ’98] [Rocca-De Zan-Monti Guarnieri-Tebaldini, ENVISAT Symp. ’07]
CRLB Sample Curves Double scatterer distance in height: 2 resolution units Relative motion: 0.7 mm/yr 4D: l.o.s. velocity 4D: height 4D: height 4D - CRLB on height estimation - No temporal decorrelation 1 10 1 antenna 2 antennas Vertical height standard deviation (m) 0 10 NO temporal [Fully ideal] decorrelation NO ATMOSPHERE -1 10 -2 10 -5 0 5 10 15 20 25 30 Signal-to-Noise ratio (dB) 2 antennas per pass Single track per pass Baseline separation: 150 m Limited advantage in the height precision limit, but gain in the SLL High gain expected in real cases with HCRLB for 4D: work in progress miscalibration residuals (atmosphere)
Conclusions In this work, we have summarize the achievements of 3D/4D SAR imaging � with satellite long-term data. The presented results demonstrate that urban scatterers can be separated � in the elevation/velocity domain by multi-dimensional imaging. By means of numerical tests and analytical bounds we have investigated � the potentialities of SAR tomography with satellite clusters. Future systems (e.g. COSMO-Skymed) or cooperative satellite formations (CartWheel, Pendulum, e.g. Tandem-X, ASI Sabrina) are expected in the future to collect high resolution data with lower temporal separation, or simultaneously. Thus, the accuracy and performance are expected to increase.
Recommend
More recommend
