Fundamentals of Remote Sensing: the Imaging RADAR System

Transcription

1 INSIS Fundamentals of Remote Sensing: the Imaging RADAR System Notions fondamentales de télédétection : le RADAR imageur Gabriel VASILE Chargé de Recherche CNRS gabriel.vasile@gipsa-lab.grenoble-inp.fr 1 Why use RADAR for remote sensing? Controllable source of illumination sees through cloud and rain, and at night Images can be high resolution 1 m 10 m with spaceborne platforms 0.1 m 0.2 m with airborne platforms Different features are portrayed or discriminated compared to visible sensors Some surface features can be seen better in radar images ice, ocean waves soil moisture, vegetation mass man-made objects, e.g. buildings geological structures 2

2 Singapore Harbour Enlargement of central ship 3 Antarctic Mapping The Erebus ice tongue 4

3 Help! A huge oil slick is approaching the coast! Mediterranean Sea 5 Oil Spill Monitoring: the southern Vietnamese coast CNES 2000 ESA 2000 Note that dark zones directly at the coast line are caused by wind shadow, not oil! 6

4 Deforestation Mapping : optical and radar sensors Optical and Radar scenes of forest clear cutting. 7 What is RADAR? Radar GRAVES (emitting site) RADAR is an acronym for RAdio Detection And Ranging Armée de l Air, 2005 Radar GRAVES (reception site) 8 3 primary functions: transmits microwave (radio) signals towards a scene receives the portion of the transmitted energy backscattered from the scene observes the strength (detection) and the time delay (ranging) of the return signals provides its own energy source active remote sensing system

5 Transmission EM spectrum of the Earth s atmosphere 9 Electromagnetic Spectrum All electromagnetic waves propagate at the speed of light X-rays, visible light, radio waves Described by variations in their electric and magnetic fields Characterized by polarization, and by frequency or wavelength (inversely proportional to frequency) Radar remote sensing the microwave portion of the electromagnetic spectrum, from a frequency of 0.3 GHz to 300 GHz, or in wavelength terms, from 1 m to 1 mm. 10

6 Relative Size of Microwave Wavelengths Microwave frequencies have been arbitrarily assigned to bands identified by letters: X-band: from 2.4 to 3.75 cm (12.5 to 8 GHz). Widely used for military reconnaissance and commercially for terrain surveys C-band: from 3.75 to 7.5 cm (8 to 4 GHz). Used in many spaceborne SARs, such as ERS-1 and RADARSAT. S-band: from 7.5 to 15 cm (4 to 2 GHz). Used in Almaz. L-band: from 15 to 30 cm (2 to 1 GHz). Used on SEASAT and JERS-1. P-band: from 30 to 100 cm (1 to 0.3 GHz). Used on NASA/ JPL AIRSAR. Radars operating at wavelengths greater than 2 cm are not significantly affected by cloud cover, however, rain does become a factor at wavelengths shorter than 4 cm. 11 What is Synthetic Aperture Radar (SAR)? A side-looking radar system which makes a highresolution image of the Earth s surface (for remote sensing applications) As an imaging side-looking radar moves along its path, it accumulates data. In this way, continuous strips of the ground surface are illuminated parallel and to one side of the flight direction. The across-track dimension is referred to as range. Near range edge is closest to nadir (the points directly below the radar) and far range edge is farthest from the radar. The along-track dimension is referred to as azimuth. Digital signal processing is used to focus the image and obtain a higher resolution than achieved by conventional radar 12

7 Principle of Synthetic Aperture Radar Temporal domain: signal processing (modulation code, correlation) signal echo target detection propagation (delay) time range positioning frequency shift (Doppler) range velocity measurement 13 Principle of Synthetic Aperture Radar Temporal domain: signal processing (modulation code, correlation) signal echo target detection propagation (delay) time range positioning frequency shift (Doppler) range velocity measurement Spatial domain: antenna processing (beam forming, antenna pointing) Separation between receiving beam directions angular positioning 14

8 Synthetic Aperture Radar: temporal domain Emission: short baseband signal Emission: frequency ramp (chirp) over B Good echolocation inter-correlation: reception - emission Power transmission! correlation peak 1/B range resolution c/2b ERS-1 satellite B = MHz ResR= 9.6 m 15 Real Aperture Radar: spatial domain Antenna processing : distance radar-target Angular resolution Resψ = 1.22 λ/l Azimuth resolution ResA = R Resψ ERS-1 satellite λ= 5.66 cm, H = 785 Km, θ= 230 mean distance R = 880 Km ResA = 6 Km!!!! 16

9 Synthetic Aperture Radar: spatial domain Antenna processing : synthetic aperture Principle: use the flight direction for a large, virtual synthetic antenna target observed over several emitted chirps 2 types of azimuth focusing algorithms: temporal (beam forming): sum up the received echoes (i.e. different range acquisitions) by compensating their phase delays with respect to the CPA (Closest Point Approach) spectral domain: the match filter ERS-1 satellite : Res A-SAR 6 m 17 Synthetic Aperture Radar: spatial domain Antenna processing : synthetic aperture Principle: use the flight direction for a large, virtual synthetic antenna target observed over several emitted chirps 2 types of azimuth focusing algorithms: temporal (beam forming): sum up the received echoes (i.e. different range acquisitions) by compensating their phase delays with respect to the CPA (Closest Point Approach) spectral domain: the match filter ERS-1 satellite : Res A-SAR 6 m 18

10 Synthetic Aperture Radar: spatial domain Antenna processing : synthetic aperture Principle: use the flight direction for a large, virtual synthetic antenna target observed over several emitted chirps 2 types of azimuth focusing algorithms: temporal (beam forming): sum up the received echoes (i.e. different range acquisitions) by compensating their phase delays with respect to the CPA (Closest Point Approach) spectral domain: the match filter ERS-1 satellite : Res A-SAR 6 m 19 Synthetic Aperture Radar: spatial domain Antenna processing : synthetic aperture Principle: use the flight direction for a large, virtual synthetic antenna target observed over several emitted chirps 2 types of azimuth focusing algorithms: temporal (beam forming): sum up the received echoes (i.e. different range acquisitions) by compensating their phase delays with respect to the CPA (Closest Point Approach) spectral domain: the match filter ERS-1 satellite : Res A-SAR 6 m 20

11 Synthetic Aperture Radar: spatial domain Antenna processing : synthetic aperture Principle: use the flight direction for a large, virtual synthetic antenna target observed over several emitted chirps 2 types of azimuth focusing algorithms: temporal (beam forming): sum up the received echoes (i.e. different range acquisitions) by compensating their phase delays with respect to the CPA (Closest Point Approach) spectral domain: the match filter ERS-1 satellite : Res A-SAR 6 m 21 Synthetic Aperture Radar: spatial domain Antenna processing : synthetic aperture Principle: use the flight direction for a large, virtual synthetic antenna target observed over several emitted chirps 2 types of azimuth focusing algorithms: temporal (beam forming): sum up the received echoes (i.e. different range acquisitions) by compensating their phase delays with respect to the CPA (Closest Point Approach) spectral domain: the match filter ERS-1 satellite : Res A-SAR 6 m 22

12 SAR focusing: ERS-1, Chamonix valley, pixels azimuth range ERS-1 RAW DATA European Space Agency 23 SAR focusing: ERS-1, Chamonix valley, pixels azimuth range ERS-1 RAW DATA European Space Agency 24

13 SAR focusing: ERS-1, Chamonix valley, pixels azimut range Range focusing RAR amplitude image SYTER, Telecom ParisTech * = 25 SAR focusing: ERS-1, Chamonix valley, pixels azimut range Azimuth focusing Range focusing SAR amplitude image SYTER, Telecom ParisTech 26

14 SAR focused SLC image 27 Multi-look processing 28

15 29 Synthetic Aperture Radar: geometrical distortions Range re-sampling foreshortening Def: Foreshortening in a radar image is the appearance of compression of those features in the scene which are tilted toward the radar. It leads to relatively brighter appearance of these slopes. 30

16 Synthetic Aperture Radar: geometrical distortions Range re-sampling foreshortening layover Def: Layover occurs when the reflected energy from the upper portion of a feature is received before the return from the lower portion of the feature. In this case, the top of the feature will be displaced, or laid over relative to its base 31 Synthetic Aperture Radar: geometrical distortions Range re-sampling foreshortening layover shadow Def: Radar shadows in imagery indicate those areas on the ground surface not illuminated by the radar. Since no return signal is received, radar shadows appear very dark in tone on the imagery. 32

17 Synthetic Aperture Radar: geometrical distortions 33 SAR geometry: ERS-1, Chamonix valley 34

18 Synthetic Aperture Radar: geometrical distortions Optical image multi-spectral, SPOT-2 SAR amplitude ERS-1, descending pass Mont-Blanc, 3D Lat/Lon geographical projection, WGS Synthetic Aperture Radar: geometrical distortions SAR amplitude ERS-1, descending Optical airborne image multi-spectral Mont-Blanc, 3D Range/Azimuth radar projection 36

19 Introduction to Speckle Image variance or speckle is a granular noise that inherently Gives a single look image a grainy, salt and pepper appearance Occupies a wider dynamic range than the scene content itself Cameroun Kourou - Guyanne 37 What is Speckle? - (1/2) Def: Speckle is coherent interference of waves scattered from terrain elements observed in each resolution cell. An incident radar wave interacts with each element of the surface and surface cover to generate scattered waves propagating in all directions. Those scattered waves that reach the receiving antenna are summed in direction and phase to make the received signal. The scattered wave phase addition results in both constructive and destructive interference of individual scattered returns and randomly modulates the strength of the signal in each resolution cell. 38

20 What is Speckle? (2/2) Addition of backscatter from a collection of scatterers produces random constructive and destructive interference Constructive interference is an increase from the mean intensity and produces bright pixels. Destructive interference is a decrease from the mean intensity and produces dark pixels. Reducing these effects enhances radiometric resolution at the expense of spatial resolution. 39 Speckle Suppression Speckle results from a coherent (phase included) process. Speckle can be reduced by incoherent (amplitude or power) processes. Speckle reduction (or smoothing) necessarily reduces the resolution (increases the resolution cell size) of single channel SAR data. Two basic linear processes: Multi-look - divides the signal into minimally overlapped frequency bands, processes each to a reduced resolution image, registers these, detects and adds the detected images. Averaging - detects the full resolution image, performs local averaging and re-sampling processes to create reduced resolution, reduced speckle images. For distributed targets both processes are equivalent. 40

21 Why averaging or sub-band addition? Joseph Goodman s speckle model or fully developed speckle (Random Walk) Received signal: sum of N elementary targets randomly distributed within the resolution cell Hypotheses: large N (N>>5) Ρi are independent and identically distributed (i.i.d.) ϕi are independent and uniformly distributed on [0,2π] S : (Re,Im) zero mean Gaussian complex circular vector variance σ2 proportional to the backscattering coefficient σ0 Attention: valid for homogeneous rough surfaces : σ0 >> λ/ (8 cosθ) 41 Fully Developed Speckle Model (1/3) 42

22 Fully Developed Speckle Model (2/3) Central Limit Theorem : for large N Re{E} and Re{E} are asymptotically zero-mean Gaussians with variance σ2 Re{E} and Re{E} independent Change of variables for I 0 and -π φ π and 0 elsewhere 43 Fully Developed Speckle Model (3/3) Phase probability density function for -π φ π and 0 elsewhere uniform distribution Intensity probability density function for I 0 and 0 elsewhere negative exponential distribution 44

23 Fully Developed Speckle: single-look amplitude R = 2σ2 : radar reflectivity (Radar Cross Section if calibrated) amplitude A= S Rayleigh probability density function (pdf) Mean Variance proportional to the mean Variation coefficient 45 Fully Developed Speckle: single-look intensity Intensity A= S 2 exponential probability density function (pdf) λ=1/r 1st 2nd equal and order moments µi=σi=r Variation coefficient ϒI=σI / µi=1 46

24 Fully Developed Speckle: multi-look intensity Intensity Gamma probability density function θ=r k=1/l Mean: Variance: Variation coefficient: 47 Fully Developed Speckle: multi-look amplitude Amplitude Nakagami-Rayleigh probability density function Mean: Variance proportional to the mean Variation coefficient: µ=1 L=0.5,1, 2,3,5 48

25 Multi-look intensity: parameter estimation (1/3) Estimation of radar reflectivity R with p(i R, L) use N i.i.d. (independent identically distributed) temporal samples I1(t1),, In(tn),, IL(tL) stochastic process Rˆ = Ψ{I(t ),...,I(t )} wide-sense stationary, ergodic process 1 L Problem: only one temporal realization of the stochastic process! Solution: use N i.i.d. spatial samples In-1(i-1,j-1), In(i,j), In+1(i+1,j+1), stochastic process ˆ R = Ψ{...,In 1 (i 1, j 1),In (i, j),in +1 (i + 1, j + 1),...} wide-sense stationary, ergodic process 49 Multi-look intensity: parameter estimation (2/3) Objective: find the estimator R=Ψ(..) Maximum Likelihood 1. Likelihood function L(R I,L) 2. Observe L i.i.d. samples L(R I1, IL) = Πi=1L L(R Ii,L) 3. Compute the log-likelihood LL(R I1, IL) 50

26 Multi-look intensity: parameter estimation (3/3) Objective: find the estimator R=Ψ(..) Maximum Likelihood 4. Minimization over R partial derivative with respect to R 5. The maximum likelihood estimator (MLE) of R is the mean estimator 3. Properties: asymptotically unbiased, i.e., its bias tends to zero as the sample size increases to infinity asymptotically efficient, i.e no asymptotically unbiased estimator has lower asymptotic mean squared error than the MLE asymptotically normal, i.e. as the sample size increases, the distribution of the MLE tends to the Gaussian distribution 51 Speckle Suppression Speckle results from a coherent (phase included) process. Speckle can be reduced by incoherent (amplitude or power) processes. Speckle reduction (or smoothing) necessarily reduces the resolution (increases the resolution cell size) of single channel SAR data. Two basic linear processes: Multi-look - divides the signal into minimally overlapped frequency bands, processes each to a reduced resolution image, registers these, detects and adds the detected images. Averaging - detects the full resolution image, performs local averaging and re-sampling processes to create reduced resolution, reduced speckle images. For distributed targets both processes are equivalent. 52

27 Speckle Suppression: mean filter (boxcar) Principle: Intensity at each sample interval in the image is replaced by the mean of pixel values in a moving window surrounding the sample. The box or mean filter preserves well the radiometry blurs textured areas 53 Speckle Suppression: median filter Principle: Intensity at each sample interval in the image is replaced by the mean of pixel values in a moving window surrounding the sample. The median filter assigns the window median value to each sample preserves texture information better modifies the radiometric information of homogeneous areas, and does not preserve point target signature NOT recommended for radar imagery! 54

28 Case study: classification Image classification categorizes image pixels into classes producing a thematic representation. Classification performed on single or multiple image channels to separate areas according to their different scattering or spectral characteristics. Classified data can be used in thematic maps, imported into a GIS or can be further incorporated into digital analysis. Thematic maps provide an interpretable summary of classes enabling analysts to associate detection capabilities of SAR imagery with terrain features. Digital image classification procedures are differentiated as being either supervised or unsupervised (clustering). 55 Unsupervised classification Unsupervised classification does not require training areas or analyst's knowledge of area Creates natural groupings present in the image values Values with similar grey levels are assumed to belong to the same cover type Analyst must determine the identity of the computer derived spectral clusters Principal clustering algorithms include: K-means clustering ISODATA clustering Fuzzy K-means clustering 56

29 K-means clustering of SAR data 1. Compute class centers: Give the initial partition (e.g. random) For each class ω, compute the initial class center Iω Iω = 1 Nω Nω I n n=1 2. Distance minimization: in each pixel (i,j) consider N spatial samples (BN7) 1. For each class ω, compute the distance measure D(I(i,j), Iω) 2. Re-assign (i,j) to the class with minimum distance measure 3. Re-compute class centers: 1. The refined class centers are computed using the partition from STEP-2 4. Test STOP condition: if NO go to STEP Total number of pixels switching classes Pre-specified number of iterations 57 Unsupervised classification: urban environment 58

30 59 Bibliography H. Maître, Traitement des images de RSO, Hermes Sciences Publications, 2001 Floyd F., Sabins JR., Remote Sensing, Principles and Interpretation, 2nd edition, Library of Congress Cataloging in Publication Data, 1986 L. Lliboutry, Sciences Géométriques et Télédétection, Masson, 1992 C. Elachi, Introduction to the Physics and Techniques of Remote Sensing, Wiley Series in Remote Sensing, 1987 Tutoriel du Centre Canadien de Télédétection (CCT): Trouvé E., Imagerie Radar à Synthèse d Ouverture, cours ETASM, Université de Savoie, 2004 Tupin F., Filtrage et extraction de caractéristiques sur les images RSO, cours ISAT, TELECOM ParisTech,