POLARIMETRIC SIGNATURES IDENTIFICATION FOR DIFFERENT FEATURES IN RADARSAT-2 POLSAR IMAGE : A CASE STUDY OF HALAYIB AREA , EGYPT

In fully polarized SAR (PolSAR) data the returned signal from a target contains all polarizations. More information about this target may be inferred with respect to single-polarization. Distinct polarization separates targets due to its different backscattering responses. A Radarsat-2 PolSAR image acquired on December 2013 of part of Halayib area (Egypt) was used in this study. Polarimetric signatures for various features (Wadi deposits, Tonalite, Chlorite schist, and Radar penetrated areas) were derived and identified. Their Co-polarized and Cross-polarized signatures were generated, based on the calculation of the backscattered power at various ellipticity and orientation angles. Graphical 3D-representation of these features was provided and more details of their physical information are depicted according to their different polarization bases. The results illustrate that polarimetric signatures, obtained due to factors like surface roughness, dielectric constant and feature orientation, can be an effective representation for analyzing various features. The shape of the signature is significant and can also indicate the scattering mechanisms dominating the features response.


INTRODUCTION
Polarimetric SARs provide significantly more data relative to conventional radars that record backscatter only at the linear polarizations.They allow measurement of the physical characteristics by using scattering mechanism between electromagnetic (EM) wave and the targets (Ulaby and Elachi, 1990) and (Zebker and van Zyl, 1991).They have been also successfully employed to classify and separate a wide range of terrain types.Fully polarimetric radars record the complete four coherent channels (HH, VV, HV, and VH) and retain the phase information.The "H" indicates horizontal and "V" indicates vertical transmit or receive polarization.The phase differences can result from a time delay when the phase velocity of H and V waves differs within the target.The total power is the sum of the power recorded for each of the linear polarizations (HH, VV, HV, and VH) (McNairn et al. 2001).
Polarimetric signature plot is a general approach to visualize the signature that captures many scattering characteristics of the ground cover targets.It is a 3D-representation of polarimetric information in various polarization bases (Jafari et al. 2014).It is usually displayed assuming: the identical transmit and receive polarizations (co-polarized) and the orthogonal transmit and receive polarizations (crosspolarized) of the wave intensity at all possible ellipticity and orientation angles.The shape of these plots is significant and can indicate the scattering mechanisms (surface, doublebounce, or multiple/volume) dominating the target response.Ellipse geometric elements are two dimensions, and target response is the third dimension represented in three dimensional coordinate system (Van Zyl et al. 1987).The shape of the Polarization signatures of the same target observed in different time should resemble each other in general if there are no changes; otherwise, they should be different.In some researches (Durden et al, 1989), (De Grandi et al, 2003), (Nunziata et al, 2011), and (Jafari et al. 2014).ithas been used as a tool for analysis and assessment of various targets.
Pedestal height is an indicator of the presence of an unpolarized scattering component, and thus the degree of polarization of a scattered wave.It can be derived and visualized on the three dimensional polarization signature plots generated from fully polarimetric data (Nunziata et al, 2011).The minimum intensity indicates the pedestal height of the polarization signature.The co-polarization pedestal height is the ratio of the maximum to the minimum received intensity when the polarizations of the transmitting and receiving antenna are the same (McNairn et al. 2001).Signatures with significant pedestals are typical of targets that are dominated by volume scattering or multiple surface scattering.(Van Zyl 1989) and (Ray et al. 1992) found that pedestal height was related to surface roughness with increases in roughness resulting in higher pedestals.

AQUIRES DATA AND THE STUDY AREA
A Radarsat-2 PolSAR image acquired on December 2013 was used in conducting this study.It was delivered as a Single-Look Complex (SLC) Standard Quad Polarization, Q6 in compressed format.The major characteristics of the image are depicted in Table 1.The study site selected is part of Halayib area located in south eastern desert of Egypt, with coordinate of 22 o 29ʹ to 22 o 09ʹ N, and 35 o 43ʹ to 36 o 03ʹ E, as shown in figure 1.A variety of different features were considered in this study.These include (Wadi deposits, Chlorite schist, Tonalite, and Radar penetrated areas).Reference data was collected from geological maps and ETM-8 images to verify the identification of these features.The Software used was the freeware Polarimetric SAR Data Processing and Educational Toolbox; (PolSARpro).
Table 1.Major Characteristics of the Used Radarsat-2 PolSAR Image

METHODOLOGY
In order to interpret and retrieve the feature information of polarimetic SAR data, pre-processing is of critical importance (Lee, et al. 2001).The first step is the generation of 2 × 2 scattering matrix [S] that measures the complete information of the surface features.This is followed by deriving the 3 × 3 coherency matrix [T3] and the polarimetric parameters.Once the scattering matrix and the covariance matrix are known, one can synthesize the received power for any transmit and receive antenna polarizations.Finally, speckle filtering and geometric correction (geo-referencing) are calculated for interpreting the image correctly.After completion of this phase, the coherency matrix is decomposed based on the Pauli basis for deriving the polarimetric signature for each surface feature on the study area.Figure 2 shows the flowchart of the different preprocessing steps for polarimetric signature retrieval.
Figure 2. Flowchart of the different pre-processing steps for polarimetric signature retrieval.
Figure 1.The study area (yellow rectangl

The Scattering Matrix
The original data is in Single Look Complex (SLC) format.
When the incident radar signal interacts with the earth feature on horizontal or vertical wave, the backscatter of the radar signal is the contribution of both vertical and horizontal wave.Therefore, the reflected backscatter can be represented by scattering matrix as given in equation 1: the co-polarized information while off-diagonal elements represent the cross-polarized information (Nandan, 2012).

Coherency Matrix
Scattering matrix is used to represent the backscatter of the coherent or pure target like urban area.In contrast, the natural target which partially polarized waves is very difficult to be realized using scattering matrix.To describe the distributed scatters, the second order matrices are used.The second order matrices are derived from the scattering matrix.In case of reciprocity condition in which   =   then the vectorized format of the scattering matrix is given in form of lexicographic basis and Pauli basis (Nandan, 2012).
In case of Pauli format: Where Kp = Pauli vector By multiplying this vector with its complex conjugate transpose the coherency matrix T3 = K P K P * is obtained: (3)

Speckle Filter
Speckle appearance in radar images is caused by the coherent interference of waves reflected from many elementry scatters (Lee and Pottier, 2009).Speckle can be reduced using multilook observations, which can be achieved during the image construction, or a speckle-reduction filter performed by the user.In order to achieve optimal speckle reduction in imagery refined lee filter was used.It was used since it has proven to be good in preserving polarimetric information for distributed targets.We tested different filter sizes and the best results were achieved by using a 7 × 7 filter.It is based on statistical correlation between channels without introducing cross talk (Niu et al. 2011), (Salehi et al. 2013).

Geometric Correction
The most significant step in SAR data pre-processing is the geometric correction.The original measure of SAR system is the slant range so, the image is recorded in slant range system (Lee and Pottier, 2009).With the slant range the image can't be visually interpreted because each pixel is compressed and can't also be display with the correct size.So, it needs to be converted into ground range.Geometric correction transfers the slant range image to ground range.The digital elevation model (DEM) of the study area was applied during performing the geometric correction.The study area is generally flat so, the terrain effects of layover and shadowing are neglected.MapReady tool has been used in conducting this part.It is a Remote Sensing Tool kit developed by Alaska Satellite Facility and embedded in the PolSARpro software.

Feature Extraction
The main objective of the decomposition of the matrix representation (e.g.coherency or covariance matrices) is extracting parameters that carry information about the structural and compositional contents of the ground target or land cover from the measured backscatter.The matrix can be first order (e.g. the scattering matrix in equation 1) or second order (e.g. the coherency matrix in equation 3).The most commonly-used decomposition is that of the scattering matrix which is known as Pauli decomposition.It decomposes the scattering matrix for mono-static case into three components for studying the surface properties which are represented as single-bounce (S HH + S VV ), double-bounce (S HH -S VV ), and volumetric ( S HV ) scattering mechanisms (Huynen, 1965), (William, 2012).These components are represented by the Pauli vector as in equation 2.

Pauli decomposition
The Pauli decomposition parameters are the elements included in the vector of equation 4. The first, second and third element in the vector represent the single-bounce, double-bounce and volume scattering, respectively.This decomposition is the most common and more appropriate for coherent targets (with identifiable structures) compared to other coherent decomposition methods.The Pauli decomposition is the most effective and useful for exposing natural targets, but not ideal for highlighting man-made targets (Zhang et al. 2008).The scattering matrix [S] can be written as: Where ∝= (S HH + S VV )/√2 β = (S HH − S VV )/ √2 γ = √2 S HV .Using Pauli decomposition, often α, β, and γ components are represented as blue, red, and green respectively for visual interpretation.The polarization color composite of the used image in Pauli basis is displayed in figure 4.

Polarimetric signature
The polarimetric signature describes the scattering coefficient as a function of any assumed transmit and receive antenna polarization states (linear, circular, and elliptical).It allows measure the variation of the scattering coefficient with polarization so that different targets show different polarization signatures (Arai, 2011).Although many targets can produce similar plots, the shape of the plots as well as the pedestal height, provide clues about the type of scattering dominant from the target.The polarimetric signatures are very sensitive to the orientation of the target relative to the radar line of sight (Schneider et al. 2005).The angle of the semimajor axis with the horizontal axis (x-axis) is the orientation angle (ψ o ) ranging from 0 o to 180 o .Ellipticity defines the oval shape of the ellipse shown as (χ o ) as depicted in figure 3. Linear polarizations have an ellipticity angle of 0 o , while circular polarizations have ellipticity angles of 45 o .Although all orientations are represented in the plot, the commonly used linear polarizations have orientation angles of 0 o (H) or 90 o (V) (McNairn et al. 2001).The polarization signature σ o is represented by the following equation 5: Where K = constant.Jr, Jt = Stokes vectors at receiver, transmitter, respectively.χ ,ψ = ellipticity and the orientation angles of the electric field vector.
With the help of the geological map and the ETM-8 image, we choose areas within the Radarsat-2 image for different types of terrain, as illustrated in figure 4. The polarization signature for each resolution element (pixel) represents the sum of the polarization signatures of each object in this pixel.

Figure 3 .Figure 5 .
Figure 3. Definition of polarization signatureThe signatures can be computed on a pixel basis or average over a region that captures the particular feature.Single pixel basis was used for the generation of the various features polarization signatures ((a) Wadi deposits, (b) Chlorite schist, (c) Tonalite, and (d) Radar penetrated areas).Co-polarization and Cross-polarization signature plots were extracted for these pixels.The following figures (5, 6, 7 and 8) display the calculated polarimetric signatures (backscattered power) of these major features normalized to the intensity range 0-1 in lin format (Mesh representation).