Radar Interferometry to Generate 3D Omar Hadj Sahraoui Engineer in Remote Sensing Division Searcher in SAR Image Team O. Hadj Sahraoui Division de Télédétection Centre National des Techniques Spatiales 01 Avenue de la Palestine B.P.13 Arzew 31200, Oran, Algérie Tel. : 00 213 71597416 – Fax :00 213 41473665 sahraoui_omar1@yahoo.fr B. Hassaine Division de Télédétection Centre National des Techniques Spatiales 01 Avenue de la Palestine B.P.13 Arzew 31200, Oran, Algérie Tel. : 00 213 71597416 – Fax :00 213 41473665 hassaineb@cnts.dz C. Serief Division de Télédétection Centre National des Techniques Spatiales 01 Avenue de la Palestine B.P.13 Arzew 31200, Oran, Algérie Tel. : 00 213 71597416 – Fax :00 213 41473665 se_chaf@yahoo.fr ABSTRACT : The availability of technical public domain software is important for a fast acceptance and integration of a technique in other disciplines, enabling scientists to gain experience with new techniques at low cost. Synthetic aperture radar (SAR) interferometry (InSAR) is an imaging technique for measuring the topography of a surface, its changes over time, and other changes in the detailed characteristics of the surface. By exploiting the phase of the coherent radar signal, interferometry has transformed radar remote sensing from a largely interpretive science to a quantitative tool, with applications in cartography, geodesy, land cover characterization, and natural hazards. Radar interferometry using freely available software has matured to a stage in which data from different sensors can be routinely processed to interferometric products. SARscape is one of the latest contributions to public-domain radar interferometry software. It is designed as a specialized software package for processing of SAR/InSAR data, and perfectly complements ENVI's functionality for analyzing and visualizing remote sensing data of any kind. More specifically, SARscape for ENVI is a modular set of ENVI (Environment for Visualizing Images) functions for processing of spaceborne remote sensing data from Synthetic Aperture Radar sensors. Currently, the systems ERS-1/2, JERS-1, RADARSAT-1 and ENVISAT ASAR are fully supported by SARscape for ENVI. In this paper, we present the feasibility of the SARscape radar interferometric software to create ERS interferometric products such as DEMs and deformation maps. A stepwise description of the creation of an ERS DEM of the Oran area in Algeria is presented. 1. Theoretical description of preparation to the interferometry: 1.1. Geographical framework  Fig.1 : Geographical Framwork The zone of interferometry study is given on a small zone of Oran ,it is in the North-West of Algeria on the area of Oran, in particular with its junction with the Verdon. It is delimited in North by the sea, of the large SEBKHA in the south and the mountains in the West and East. From a strictly topographic view point, a characteristic of the zone and strong vegetable cover (Forest) on the sites mountainous and in the south as the East, the landscape does not present a marked relief (uneven the 200m do not exceed). 1.2. Selection of an interferometric couple: The first requirement is the availability of two SAR images in complex form. Complex SAR data refer to a set of data that has a real (cosine) and an imaginary (sine) component. The two values combine as vectors to provide the overall phase and intensity of a wave. Both these components of backscattered signals are measured by the SAR sensor onboard the satellite. This provides two resulting data streams, namely ‘I’ (representing In-phase / intensity / cosine component) and ‘Q’ (representing Quadrature / phase / sine component). The selection of the images is made on the basis of baseline length and the time period between two image acquisitions. Depending upon the application and the spatial resolution of the data, the baseline length can be chosen. For example, in the case of ERS–1 and 2, the baseline may be taken as 150 to 300m for topographic applications, 30 to 50m for surface change detection and up to 5m for surface feature movement studies such as crustal deformations, lithospheric movements, movement of glaciers etc. Also, the time gap between two passes of satellite may not be kept large as there may be some changes in the scene that may lead to temporal de-correlation. However, the temporal de-correlation, in the case of ERS-1 and 2 may be taken care of by tandem operation of two satellites at a small temporal resolution of as low as one day. The selected raw data are then processed to convert SAR signals to image products like, Single Look Complex and GTC Geocoded Terrain Corrected with the help of DEM. This processing requires knowledge about the precise orbit and calibration parameters such as time reference and intervals of each image, and the chosen spatial and temporal resolutions of the images. 1.2.2. Topographic data For the software most powerful to the realization of the interferogramme, use differents DEM of differents precisions for exemple GTOPO 30. GTOPO30 is presented in the form of a grid of geographical co-ordinates associated ellipsoid WGS84; its reference in altitude is the geoid. Its 30 seconds resolution of arc provides a point all the 900m approximately in latitude. 2. SAR Interferometry for DEM Generation: Two SAR images are combined to produce a SAR Interferogram to reveal information about the third dimension (elevation) of the object and to measure small displacements of objects between the two image acquisitions. An interferogram is an image acquired by making the phases of two SAR images of the same terrain to interfere. Thus, after registration, the complex interferograms are formed by multiplying each complex pixel of the first image by the complex conjugate of the same pixel in the second image. The interferogram thus generated is a complex image itself. The intensity of the interferogram is a measure of cross correlation of the images. A careful observation of the fringes in Fig.4 reveals that closer are the fringes, more are the topographical changes or height variations. 2.1. The Registration The chock of the interferometric geometry is one of the stages most significant and delicate in the process of generation of DEM. Of its quality the precision depends directly on the base line. However, the planimetric and altimetric errors obtained on a DEM are primarily due to an insufficient precision on the base line. It is thus significant to precisely reconstitute the orbital trajectories of the satellites in order to know their position perfectly. To readjust the two images of the same scene makes it possible to handle them in the same geometry. One of them is taken as reference; one defines it as main image. By opposition, the other image is called image slave. 2.1.1. Geometric registration Master/DEM At the time of this first stage, SARSCAPE fixes the main image compared to the DEM used with a precision of the order of a fraction of pixel of the DEM. In fact, this last fact the object of a radiometric simulation according to the geometry of catch of sight of the image radar defined in a file coarse descriptor. The image of the DEM projected in geometry radar is then similar to the main image, with the inaccuracies on the orbital parameters NR and to near. It is then possible to carry out a correlation between these two images by pairing of homologous groups pixellic, for which one obtains a peak of correlation. This module of the software provides three imagettes representing the shift in distance, azimuth and the rate of correlation between the main image and simulation. From these shifts, SARSCAPE generates a new file more precise descriptor by correcting the values of NR and to. The precision of this chock controlled by the orbitography and the DEM depends mainly on two criteria. On the one hand, the step of the grid of the DEM must be sufficiently small. With GTOPO30, whose step is 900m, it is necessary to use points of supports. In addition, it is preferable to have on the image a well marked relief, from where interest of the total treatment of great surfaces, the more so as the points of supports are relatively rare on the scenes radar. 2.1.2. Registration image Master/ slave: In cases where multiple image data sets cover the same region, it is necessary that pixels in different images correspond so that pixel-by-pixel comparisons can be carried out. Spatial registration may be necessary, and also resampling, in cases where pixel sizes vary. Coregistration is carried out automatically, based on maximising correlation in a number of windows. 2.1.3. Orbit refinement: The role of the orbital parameters to and NR is very important since it is them which will allow the passage of the satellite reference mark (azimuth, distance)to the geographical reference mark. Thus, in the absence of precise topographic data, it can appear skews on their estimate. It is the case during the registration of the main image relative with GTOPO30, but the use of a certain number of points of supports makes it possible to control the validity of this chock. These points of supports are GPS points which should obligatorily be measured. These remarkable points can be localised on the image, their indices of line and of column (i,j) are thus known. One has then: In the same way, the knowledge of time makes it possible to calculate: i=(t-to)/PRF j=(R-NR)/ða PRF: Pulsate Repetition Frequency. d D : Radial resolution. NR : Near Arranges Now and in this stage which one must compare measurements with truths values of the file orbits. The comparison between the indices measured and the file orbits and estimated allows to determine the shifts in line and column. taking some points for example in this table:  thus that deduit: dLmoyen=-35 dCmoyen=33 These shifts of pixels dL and cd. are then retranscribed respectively in distance and time lags (dt=dL/PRF) (dR=dC.dd). By applying these shifts to the values of to and NR calculated by SARSCAPE, we take a better precision of registartion. So it’s verry interestant to integrate the new parameters t0 and NR adjusted in the file descriptor of SARSCAPE and restart again the interferometric chain. 2.2. Interferometric product The interferometric product of SARSCAPE consists of three images: an image of phase or interferogram, an image of coherence and an image of amplitude, the latter being formed by average of the amplitudes of the complex interferogram. SARSCAPE offers two geometries of exit: that of the images radar and that of the DEM. In the case of geometry radar, the pixels are gathered according to initial factors' multivue. The produced images are then superposable with the multivue of the main image. In the case of the otho-rectified geometry (DEM), the pixels are gathered by closer neighbors on the points of the DEM. The produced images are then superposable with the DEM. Thereafter, we will use the interferograms exclusively géocodés. In geometry radar, the interferograms obtained can be interpreted like a chart of the topographic errors of the DEM of entry translated into dephasings. Also, the topographic fringes of the interferogram calculated with IGN50 are rare even non-existent because the precision of altitudes is remarkable. On the other hand, with GTPOO30, it remains much of fringes due to the relief, because its precision is insufficient. 2.3. Elimination of the orbital fringes As we saw previously, the precision of the orbital data is insufficient on a centimetric scale wavelengths used. The compensation of the orbital fringes is thus imperfect and is translated on the interferogram by a residual gradient in the radial direction and in the azimuth direction (defect of parallelism of the orbits). The end of the treatment aims at eliminating these fringes. Thus, by simple counting of the orbital fringes, one can consider the "slope" average of the fringes and eliminate them by withdrawing from the interferogram the gradients corresponding (in distance and azimuth). This stage can be carried out manually. In this case, the interferogram produces with IGN50 is invaluable. Indeed, the gradients observed on one or the other of the interferograms are appreciably the same ones since the conditions of catch of sight are identical (if one excludes the stage of chock of the main image on the DEM). However, the fact of being able to have another DEM as IGN50 is not systematic, especially when one seeks to rebuild the initial DEM. One can then count on the automatic module of SARSCAPE which envisages the elimination of the orbital fringes on the interferogram even disturbed. Without reaching a as good estimate as manually, the software thus calculates the residual gradients in azimuth and distance. 3. Restitution of the relief: After having seen the process of formation of the interferograms, we approach here a method of restitution of the relief. 3.1. Choice of the zones of test To proceed to the realization of the relief, it is always necessary to choose several small zones with the different topographic characteristics, for which we will discuss the treatments and the effectiveness of the restitution. 3.2. Filtering According to the altitude of ambiguity employed, the interferogram can contain an excessive noise, in particular on zones with strong slopes, likely to obstruct its exploitation since the fringes are found masked. A stage of filtering thus is essential often in the data processing sequence of interferograms ROS to eliminate the noise which is added with the exploitable signal.The difficulty of filtering consists in cleaning the noise without degrading the information of phase. This noise of space or temporal origin is not always reversible. Only, one considerable share of the noise affecting the interferograms can be eliminated, which leads to information of cleaned phase which guarantees a more robust course of phases and débruitée final information . Several filter integrated on this software which one used the filter Gamma-Gamma:  Fig 2 : No Filtred and Filtred phase image 3.3. Phase unwrapping In an interferogram, the phase is only known modulos 2 p. It is thus N écessaire to determine the multiple of 2 p to add with the phase measured on each point to obtain an estimate of the real phase. The course of phase thus consists in redistributing with each pixel its absolute phase. Two constraints burden this procedure:
 Fig 3 : Phase unwrapping in 3D the phase unwrapping is the most difficult stage in the process of generation of a DEM. Ainsi, the presence of noise on the interferogram can distort the information of phase considerably. This is why it must be preceded by an effective filtering which cleans this noise. A first solution to unroll the phase consists in correcting a pixel compared to an adjacent pixel by supposing that the slope between these two points is weakest possible among all those possible defined with a margin of 2??. In practice, the interferogram is traversed starting from a pixel reference and each pixel is corrected gradually by taking account of the precedent.  Fig. 4: Interferogram unwrapping The mottled shapes of the interferogram take the aspect of smooth and compact structures after filtering.the phase unwrapping was thus made without apparent discontinuity. Our work leads has to carry out the DEM shown on the following figure.  Fig.5 : Degital Elevation Model Conclusion: Following the treatment that we imposed on the data interferometric available on the zone of study, we obtained a Digital Elevation Model which although presenting notable imperfections, reaches details much more precise than the starting DEM GTOPO 30. The structures thus formed provide space and altimetric informations, compatible with topographic applications. Reference:
|