An airborne X and L-band Synthetic Aperture Radar system was developed by the joint project of Communication Research Laboratory and National Space Development Agency of Japan from 1993 to 1996. It is installed on the airplane, Gulfstream II. The resolution is 1.5 m/3.0 m (for X/L-band) in both azimuth and range direction. The both SAR are operated with polarimeter capability. The X-band SAR has cross-track interferometry function. In this report we describe our SAR system, ground processing system and the performance of our system. Furthermore we will discuss motion compensation and interferogram quality.
This paper describes the implementation of a real-time radar data processor for an airborne research radar system: the multimode, polarimetric, interferometric Enhanced Surveillance Radar (ESR) developed by the UK Defence Evaluation and Research Agency (DERA). The computer performance available in today's shared memory multiprocessor workstation has enabled us to design and implement a radar data processor which not only achieves real-time performance but provides an uniquely flexible environment for undertaking research and development into radar processing techniques.
Ground imaging using a Synthetic Aperture Radar (SAR) requires the knowledge of the antenna trajectory with a relative accuracy of a fraction of the wavelength upon an integration time of a few seconds. This information is not always available from inertial navigation unit especially in the airborne framework and must be recovered from the radar signal itself using the techniques known as 'autofocus.' We describe in this paper an original method designed at ONERA for the registration of looks. We show how it applies to autofocusing. It improves the map-drift algorithm making it time variant thus allowing for the correction of low frequency trajectory errors, and not only velocity bias.
Recently developed Scanning Synthetic Aperture Radar (ScanSAR) instruments require new algorithms for the processing stages because the classical approaches to SAR image generation are not well suited to ScanSAR data. Therefore, other techniques must be investigated. Spectral Analysis (SPECAN) is an efficient implementation of chirp de-correlation and it is specially well adapted to ScanSAR bursts of data. It consists of deramping and a spectral analysis stage that can be implemented by several techniques. One of them is Fast Fourier Transform (FFT) that is very efficient but presents some drawbacks concerning pixel spacing. A recently proposed method is the Chirp Z-Transform (CZT) SPECAN that obtains spectrum information at arbitrary positions. CZT can be efficiently implemented by FFTs and mixing operations and it provides the desired flexibility to the user. The main contribution of this study is the analysis of SPECAN techniques for processing ScanSAR data by comparing spectral estimation methods and new multilook strategies. The new CZT-SPECAN technique is compared with the classical SPECAN approach showing its advantages avoiding interpolation steps. The new multilook strategy implemented by SPECAN also reduces the scalloping effect. The results show that the new method is very promising for obtaining fast and simple ScanSAR image generation.
The wavelet transform has become a very popular tool in signal and image processing. Over the last few years, several authors have proposed wavelet-based filters for speckle reduction in SAR*images, and the results are generally reported to be superior to those obtained with traditional statistical speckle filters. In this paper we give a thorough experimental comparison of representative filters from both categories. We show that spatially adaptive statistical filters yield better noise reduction and preservation of structures than wavelet- based methods, but that the latter have certain advantages compared to statistical filters which are not spatially adaptive.
Multi-angle, multi-polarization C-band backscattering measurements were performed over selected bare soil areas. To perform these measurements, an FM-CW radar has been designed and assembled. This device has the capability of resolving independent samples within the antenna footprint area, thus allowing range discrimination and improving the signal statistics. Two areas with different degrees of roughness and dielectric constants were selected and set up. Co-polarized backscattering coefficients were measured for incidence angles between 23 degrees and 60 degrees. To perform a model analysis of the backscattering properties, 'ground truth' data, including surface roughness profiles and soil moisture values (directly related to dielectric constant) were also collected. The 'classical' parameters, used to describe surface roughness, showed a wide spreading. This evidence and the data resulting from ground truth campaigns over many European test sites suggested an alternative description of surface roughness, based on the self-similarity (fractal) properties. The surfaces have therefore been described as fBm (Fractional Brownian Motion) processes, and their backscattering response has been theoretically modeled by a numerical simulation (in 3-D in order to also take into account anisotropy effects) in Kirchhoff approximation. The experimental data have been analyzed with both asymptotic models (IEM) considering a classical statistical description, and with the numerical simulation applied to fBm surfaces.
Proc. SPIE 3497, Inversion of electromagnetic models for estimating bare soil parameters from radar multifrequency and multipolarization data, 0000 (20 November 1998); https://doi.org/10.1117/12.331363
In this work we assess the practical feasibility of bare soil parameter estimation using multifrequency SAR polarimetric data. We use both the IEM and a semi-empirical model to simulate the radar measurements. We account for the multidimensional noise which is present in the measured polarimetric covariance matrix and compare different inversion algorithms, namely multivariate regression, maximum likelihood and minimum variance algorithms, and a neural network. The comparison are performed using different sets of radar parameters in order to assess the accuracy achievable by different radar configurations.
The paper considers the problem of detecting point-like targets in SAR imagery. We review the theory of constant false alarm rate (CFAR) detection, but note that on real SAR images CFAR performance is limited in two respects: (1) Inhomogeneities in the background cause a marked increasing in the false alarm rate. (2) As the target may only subtend a few pixels, we can only reduce the uncertainty introduced by speckle by increasing the size of the background region. We suggest that these problems can be overcome by using segmentation as a method for obtaining optimal background regions and verify this with experiment.
It is widely recognized that SAR images exhibit a fractal behavior represented by the concept of fractal dimension, which is related to an intuitive concept of surface 'roughness.' The most suited approach to compute the fractal dimension comes from the power spectra of a fractal Brownian motion: the ratio between energies at different scales is related to the persistence parameter H and, thus, to the fractal dimension D equals 3 - H. The signal-dependent nature of speckle, however, prevents from the exploitation of this property to estimate the fractal dimension of SAR images. In this paper, we propose and assess a novel method to obtain such a fractal signature, based on the multi-scale image decomposition provided by the normalized Laplacian pyramid (NLP), which is a bandpass representation obtained by dividing the layers of the LP by expanded versions of its baseband, designed to exhibit noise that is independent of the signal. Thus, by analyzing SAR image texture on multiple scale through the NLP, it is possible to highlight and assess fractal behaviors of the radar cross-section. Experiments on both synthetic and true SAR images corroborate the theoretical assumptions underlying the proposed approach.
We have recently presented optimal criteria for edge detection and edge localization in Single Look Complex (SLC) Synthetic Aperture Radar (SAR) images. By working on complex data rather than intensity images, we can easily take the speckle autocorrelation into account, obtain more accurate estimates of local mean reflectivities, and thus achieve better edge detection and edge localization than with operators known from the literature. After a review of the theoretical aspects, we here propose solutions for the practical implementation. In SLC images the Maximum Likelihood (ML) estimator of reflectivity is the Spatial Whitening Filter (SWF), which is used in both tests. It necessitates precise knowledge of the speckle correlation. We describe how it can be determined and discuss the consequences of inaccuracies. Two-dimensional edge detection can be realized with multidirectional sliding windows. The watershed algorithm permits the extraction of closed, skeleton boundaries, and thresholding of the basin dynamics efficiently reduces the number false edges. The optimal estimator of the edge position is computationally intensive, so we examine suboptimal methods which require far less multiplications. The edge localization stage can be implemented with active contours or Gibbs random field techniques. Some segmentation results are shown.
The presence of speckle, which may be modeled as a strong multiplicative noise, makes the segmentation of synthetic aperture radar (SAR) images very difficult. The usual gradient operators yield poor results, but robust operators have been developed specifically for this kind of images. From the edge strength map ('gradient image') created by such an operator, closed skeleton boundaries running through local maxima must be extracted. This can be achieved with the watershed algorithm. However, to reduce the number of false edges, the algorithm must be made less sensitive to speckle. In this article, we compare two different approaches -- watershed thresholding and basin dynamics -- and propose a new algorithm for the computation of edge dynamics. The improvement brought by basin and edge dynamics is illustrated on ERS-1 images of an agricultural zone.
The retrieval of ocean wave spectra from ERS SAR image cross spectra is addressed in order to assess their potential to estimate the thickness of thin sea ice such as frazil and pancake ice. The inversion procedure based on the gradient descent algorithm, already demonstrated for airborne SAR data, is exploited and the capability of this method when applied to satellite SAR sensors is investigated. In fact the major differences between the two imaging situations lie in the illumination geometry and azimuth integration time. The SAR- ERS SLC image acquired on April 10, 1993 over the Greenland Sea was selected as test image. A couple of windows that include open sea only and sea ice cover, respectively, were selected. The inversions were carried out using different guess wave spectra taken from SAR image cross spectra. Moreover, care was taken to properly handle negative values eventually occurring during the inversion runs. This results in a modification of the gradient descending technique that is required if a non-negative solution of the wave spectrum is searched for. Results are discussed in view of the possibility of SAR data to detect ocean wave dispersion as a means for the retrieval of ice thickness.
In the last years, both local and global analysis techniques for the effective processing of interferometric SAR data have been proposed. We developed two local approaches to eliminate inconsistencies in the measured (wrapped) phase field, based on the local configurations of phase gradients in finite windows. The first technique adopts a fixed search strategy which 'cures' isolated residue couples by an appropriate series of corrections determined a priori. A second strategy uses the generalization capabilities of a neural network, trained on a suitable number of simulated target phase fields, to add 2 - (pi) cycles to the proper locations of the interferogram. These approaches, in spite of the high dimensionality of this problem, are able to correctly remove more than half the original number of pointlike inconsistencies on real noisy interferograms. This stems from the observation that phase unwrapping is an ill-posed problem, which has to be solved globally. Hence, a global stochastic method has been implemented, based on the minimization of a functional measuring the regularity of the phase field. The optimization tool used is simulated annealing with constraints. This methodology gives excellent results also in difficult conditions. We will present some of the recent results which aim at integrating the above-mentioned methodologies into powerful processing chains optimized for operating on large IFSAR datasets from real scenes. The effectiveness of such phase retrieving methods allows the application of sophisticated and innovative remote sensing techniques, such as differential interferometry.
The proposed method for phase unwrapping is based on a global analysis of the interferometrical phase. The underlying principle is that the interferogram is partitioned such that the unwrapped-phase function on each element can be locally modelled by the mean values of the phase difference between neighboring pixels in azimuth and range directions. Using this local information and a least-squares algorithm (Gauss-Seidel relaxation), an approximate model of the unwrapped phase is then generated and tested by calculating the 'residue image' defined as the difference between the original interferogram and the model itself. If there are residual fringes, then the result must be iteratively refined applying the method to the residue image. The accuracy of the proposed estimation depends on the dimensions of the elements and the dynamic content of the phase, i.e. on the 'roughness' of the ground surface. In order to limit the influence of noise and layover on the unwrapped-phase function generation, we are working to a 'weighted' version of our algorithm. This approach introduces for each element of the partition a coefficient of confidence representing the reliability of local slope estimation.
The phase unwrapping is the key step in recovering the terrain elevations from Interferometric SAR data. Phase unwrapping deals with the problem of estimating the absolute phase from observations of its noisy wrapped values. It is an ill-posed inverse problem. We propose a Bayesian model fitting solution using a fractal prior in the form of a multiscale stochastic process.
Shuttle Radar Topography Mission (SRTM) X-band simulated interferometric raw signal pairs are used to test the performance of a new phase preserving Signum Coded SAR Processor (SCSP) for real time operations. Raw signal pairs relevant to both a canonical and a real scenario are simulated. The canonical scene consists of a pyramid and three corner reflectors. The real scene refers to the Mt. Etna area, in Sicily, Italy. Full result comparison between SCSP and conventional products is presented.
A quality assessment of InSAR topographic mapping was carried out addressing the issues of feasibility and quality in the particular case of Belgium. We first present a pre-study consisting in the selection of the most appropriate image pairs as well as an assessment of the expected DTM quality, based on interferometric baseline values and ancillary data like cloud cover, humidity, vegetation cover, relief, etc., that influence the coherence level. The quality of the DTM generated from the selected ERS-1/2 Tandem pairs is discussed by comparison with a reference DTM and put in relation with the coherence maps obtained simultaneously with the interferograms.
The synergetic use of optical (SPOT stereo images) and radar (interferometric SAR images) data for DEM generation is addressed. The paper presents a complete interferometric SAR (InSAR) procedure and investigates the potentiality of SPOT data to support such a procedure and to improve the quality of the terrain surface reconstruction using a flexible data fusion approach.
In 1996, the French research group ISIS* proposed a research initiative in the field of radar imaging. One purpose aims to study specificities of multitemporal SAR (synthetic Aperture Radar) images. This paper presents some results of research undertaken in the multitemporal workgroup. In this paper two different kinds of SAR multitemporal data have been distinguished: temporal sequences of spatial images and images containing temporal information like interferograms. To each of them, correspond different types of image processing. The multitemporal information may be used either to enhance static information (multitemporal filtering) or to study temporal evolution (change detection, temporal tracking of structures). In this paper, two different methods of quality restoration are proposed to enhance amplitude data using sequences of SAR images. Then a model for detecting and tracking environmental non-rigid structures (like the coastal line) is explained. Lastly, a segmentation method for phasimetric effects on interferograms is described.
The present work consists on the generation of a DEM using ERS satellites interferometric data over a wide area (50 X 50 Km) with an error study using a high accuracy reference DEM, focusing on the atmosphere induced errors. The area is heterogeneous with flat and rough topography ranging from sea level up to 1200 m in the inland ranges. The ERS image has a 100 X 100 Km2 area and has been divided in four quarters to ease the processing. The phase unwrapping algorithm, which is a combination of region growing and least squares techniques, worked out successfully the rough topography areas. One quarter of the full scene was geocoded over a local datum ellipsoid to a UTM grid. The resulting DEM was compared to a reference one provided by the Institut Cartografic de Catalunya. Two types of atmospheric error or artifacts were found: a set of very localized spots, up to one phase cycle, which generated ghost hills up to 100, and a slow trend effect which added up to 50 m to some areas in the image. Besides of the atmospheric errors, the quality of the DEM was assessed. The quantitative error study was carried out locally at several areas with different topography.
The mass production of SAR products and its usage on monitoring emergency situations (oil spill detection, floods, etc.) requires high-speed SAR processors. Two different parallel strategies for near real time SAR processing based on a multiblock version of the Chirp Scaling Algorithm (CSA) have been studied. The first one is useful for small companies that would like to reduce computation times with no extra investment. It uses a cluster of heterogeneous UNIX workstations as a parallel computer. The second one is oriented to institutions, which have to process large amounts of data in short times and can afford the cost of large parallel computers. The parallel programming has reduced in both cases the computational times when compared with the sequential versions.
Synthetic aperture radar (SAR) imagery is now well-known as it is actively used in various remote sensing domains such as oceanography, geology, environment surveillance, cartography, etc. In this paper, we describe a new method for computing an elevation map from a single SAR image. Our method begins by reconstructing an approximate elevation map of the imaged ground using radarclinometry. This reconstruction relies on a Lambertian assumption for the backscatter model, which, as our experiments have shown, is not particularly accurate. Therefore, once we have computed the initial estimation, we proceed to estimate a more accurate backscatter model for the imaged ground. After a brief description of the relevant reflection mechanisms, we show that the theoretical considerations are too complex to be completely and accurately modeled. We therefore present two methods for performing the backscatter estimation. The first method is based on empirical results. The second method makes use of the approximate elevation map obtained by radarclinometry. In the latter case, we subsequently compute an improved elevation map using the more accurate backscatter model.
SAR image simulation is important for a number of tasks in the design of SAR dedicated systems. We propose here a SAR texture simulation tool. Based on a statistical description of the local reflectivity and elevation of a surface, this tool generates seamless texture mapping patterns. The associated SAR texture renderer can provide several levels of realism, from flat 2-D texture mapping to 3-D unperfectly conducting surfaces back-scattering model, going through normal dithered Lambert law. The back-scattering model simulates polarizing, overlaying and shadowing effects, but not the multiple reflections. Using the range-buffer technique, the synthetic texture patterns are mapped on 3-D objects.
This paper deals with a new exact formulation using SAR interferometry approach to extract accurate terrain elevation. By using the geometric formulation relating the elevation of a ground point to the phase difference of SAR images, we obtain an exact third order expansion with Maple software. Compared with this expansion, the second order one presents an error which can be expressed in terms of fringes shift in the interferogram. As the geometric InSAR formulation is different from one author to another, we present the utility of our work, and compare our results with those obtained by Lin (1991). We demonstrate that our more general formulation yields the development adopted by Prati and Rocca for interferogram generation. We also describe the procedure for phase generation from interferometric image pair. The baseline computation depends on the choice of the path length difference that we decide to adopt. By the same way, we illustrate the effects of the earth rotundity on the map resolution accuracy.