Céline Tison

A new statistical model for Markovian classification of urban areas in high-resolution SAR images

We propose a classification method suitable for high-resolution synthetic aperture radar (SAR) images over urban areas. When processing SAR images, there is a strong need for statistical models of scattering to take into account multiplicative noise and high dynamics. For instance, the classification process needs to be based on the use of statistics. Our main contribution is the choice of an accurate model for high-resolution SAR images over urban areas and its use in a Markovian classification algorithm. Clutter in SAR images becomes non-Gaussian when the resolution is high or when the area is man-made. Many models have been proposed to fit with non-Gaussian scattering statistics (K, Weibull, Log-normal, Nakagami-Rice, etc.), but none of them is flexible enough to model all kinds of surfaces in our context. As a consequence, we use a mathematical model that relies on the Fisher distribution and the log-moment estimation and which is relevant for one-look data. This estimation method is based on the second-kind statistics, which are detailed in the paper. We also prove its accuracy for urban areas at high resolution. The quality of the classification that is obtained by mixing this model and a Markovian segmentation is high and enables us to distinguish between ground, buildings, and vegetation.

Support Vector Machine for Multifrequency SAR Polarimetric Data Classification

The objective of this paper is twofold: first, to assess the potential of radar data for tropical vegetation cartography and, second, to evaluate the contribution of different polarimetric indicators that can be derived from a fully polarimetric data set. Because of its ability to take numerous and heterogeneous parameters into account, such as the various polarimetric indicators under consideration, a support vector machine (SVM) algorithm is used in the classification step. The contribution of the different polarimetric indicators is estimated through a greedy forward and backward method. Results have been assessed with AIRSAR polarimetric data polarimetric data acquired over a dense tropical environment. The results are compared to those obtained with the standard Wishart approach, for single frequency and multifrequency bands. It is shown that, when radar data do not satisfy the Wishart distribution, the SVM algorithm performs much better than the Wishart approach, when applied to an optimized set of polarimetric indicators.

A Fusion Scheme for Joint Retrieval of Urban Height Map and Classification From High-Resolution Interferometric SAR Images

The retrieval of 3-D surface models of the Earth is a major issue of remote sensing. Some nice results have already been obtained at medium resolution with optical and radar imaging sensors. For instance, missions such as the Shuttle Radar Topography Mission (SRTM) or the SPOT HRS have provided accurate digital terrain models. The computation of a digital surface model (DSM) over urban areas is the new challenging issue. Since the recent improvements in radar image resolution, synthetic aperture radar (SAR) interferometry, which had already proved its efficiency at low resolution, has provided an accurate tool for urban 3-D monitoring. However, the complexity of urban areas and high-resolution SAR images prevents the straightforward computation of an accurate DSM. In this paper, an original high-level processing chain is proposed to solve this problem, and some results on real data are discussed. The processing chain includes three main steps, namely: (1) information extraction; (2) fusion; and (3) correction. Our main contribution addresses the merging step, where we aim at retrieving both a classification and a DSM while imposing minimal constraint on the building shapes. The joint derivation of height and class enables the introduction of more contextual information. As a consequence, more flexibility toward scene architecture is possible. First, the initial images (interferogram, amplitude, and coherence images) are converted into higher-level information mapping with different approaches (filtering, object recognition, or global classification). Second, these new images are merged into a Markovian framework to jointly retrieve an improved classification and a height map. Third, DSM and classification are improved by computing layover and shadow from the estimated DSM. Comparison between shadow/layover and classification allows some corrections. This paper mainly addresses the second step, while the two others are briefly explained and referred to already published papers. The results obtained on real images are compared to ground truth and indicate a very good accuracy in spite of limited image resolution. The major limit of DSM computation remains the initial spatial and altimetric resolutions that need to be made more precise

Feature fusion to improve road network extraction in high-resolution SAR images

This letter aims at the extraction of roads and road networks from high-resolution synthetic aperture radar data. Classical methods based on line detection do not use all the information available; indeed, in high-resolution data, roads are large enough to be considered as regions and can be characterized also by their statistics. This property can be used in a classification scheme. Therefore, this letter presents a road extraction method which is based on the fusion of classification (statistical information) and line detection (structural information). This fusion is done at the feature level, which helps to improve both the level of likelihood and the number of the extracted roads. The proposed approach is tested with two classification methods and one line extractor. Results on two different datasets are discussed.

Retrieval of building shapes from shadows in high resolution SAR interferometric images

Discontinuous objects, such as buildings, produce shadows in SAR images. Shadows are striking features which greatly help in the image understanding. Due to the high density of buildings in urban areas, shadows cover a large part of the image and provide a major hint to build a map of the city. A straightforward use of the shadows is to determine the building height from the shadow dimensions. We propose another approach here which makes use of the shadow to help in detecting the building itself when a high resolution interferogram is available. Starting from an amplitude image with very high definition and the corresponding interferogram, we model the building detection problem as an energy minimization where the interaction between a building and its shadow is taken into account. The method allows to obtain excellent detections especially for high or isolated building, despite the important noise level.

Classification of Tropical Vegetation Using Multifrequency Partial SAR Polarimetry

This letter presents a case study addressing the comparison between different synthetic aperture radar (SAR) partial polarimetric options for tropical-vegetation cartography. These options include compact polarization (CP), dual polarization (DP), and alternating polarization (AP). They are all derived from fully polarimetric (FP) SAR data acquired by the airborne SAR (AIRSAR) sensor over the French Polynesian Tubuai Island. The classification approach is based on the support vector machine algorithm and is further validated by several ground surveys. For a single frequency band, FP data give significantly better results than any other partial polarimetric configuration. Among the partial polarimetric architectures, the CP mode performs best. In addition, the DP mode shows better performance than the AP mode, highlighting the value of the polarimetric differential phase. The combination of different frequency bands (P-, L-, and C-bands) holds the most significant improvement: The multifrequency diversity adds generally more information than the multipolarization diversity. A noticeable result is the major contribution of the C-band at VV polarization (the only polarization available at C-band with the AIRSAR data set used in this letter) to the classification performance, due to its ability to discriminate between Pinus and Falcata.

Time-Frequency Analysis in High-Resolution SAR Imagery

In this paper, a time-frequency analysis (TFA) is proposed to derive the backscattering properties of each pixel in single-polarization synthetic aperture radar (SAR) images. At high resolution (HR), some backscattering variations which are linked to the scene geometry and the surface property occur during the radar acquisition. TFA permits to retrieve these variations from the synthesized images. The proposed TFA algorithm is based on a sliding bandpass filtering in the Fourier domain, from which a spectrogram featuring the range and azimuth backscattering variations is derived. The spectrograms summarize the physical properties of each pixel. From the spectrogram analysis, four target classes representing the four main kinds of backscattering behaviors observed in SAR images are defined: frequency invariant, range variant, azimuth variant, and 2-D variant. These classes can further be linked to the physical properties of the objects. An original and simple set of five features estimated from spectrograms is proposed to classify point targets into these four classes. A performance assessment of this classification is carried out, using ONERA/RAMSES X-band airborne images acquired over the city of Toulouse, France. A robustness analysis is also conducted, in order to assess the impact of incidence angle and resolution on the classification performance. Finally, results are also given for spaceborne images (TerraSAR-X spotlight images). The physical interpretation developed in airborne case appears to be also valid for metric spaceborne data. After studying the TFA on HR spaceborne images, the tradeoff between HR coupled with TFA and medium resolution coupled with polarimetric analysis is investigated. Actually, TFA represents another way of characterizing the physical mechanisms involved in image formation.

Use of the SVM Classification with Polarimetric SAR Data for Land Use Cartography

Yhis study comes within the framework of the global cartography and inventory of the Polynesian landscape. An AIRSAR airborne acquired fully polarimettric data in L and P bands, in August 2000, over the main Polynesian Islands. This study focuses on Tubuai Island, where several ground surveys allow the validation of the different results. Different decompositions, such as H/A/alpha , or based on the Pauli formalism have shown their potential for land use discrimination. In order to take into account these different parameters into a supervised classification scheme, the SVM (Support Vector Machine) method is investigated. When dealing with only the coherent matrix elements, the results show that the SVM classification gives comparative results to those obtain with Wishart classification. Results are significantly improved when adding to the coherent matrix elements, other polarimetric parameters, as H/A/alpha or the co-polarized circular polarization correlation coefficient, rhorrll, for the Support Vector definition. Finally the best results are given when merging all the parameters for P and L bands, in addition to the only VV single channel acquired in C band.

Target recognition in SAR images with Support Vector Machines (SVM)

This paper addresses object recognition problem in SAR images with SVM classifier; the work has been mainly focused on feature vector definition. Actually, each object is represented by a feature vector and SVM aims to estimate the best hyperplanes that separate classes in the feature space. Very robust definition of feature vector is proposed and tested on real data (MSTAR database). Confusion matrices prove that a very good recognition rate is reached, even for mixed incidence angles configuration.

Polarimetric Analysis of Bistatic SAR Images From Polar Decomposition: A Quaternion Approach

This paper focuses on polar decomposition, which is based on the quaternion formalism, in single-look and multilook synthetic aperture radar polarimetry. Polar decomposition is used to decompose a bistatic or monostatic polarimetric scattering matrix into a product of a Hermitian matrix (boost) and a unitary matrix (rotation). After an overview of polar decomposition principle and quaternion properties, coherent (single-look complex) and incoherent (multilook) polar decompositions are discussed. In single-look polar decomposition, we introduce the boost parameter and the rotation parameter with the purpose of classifying scattering mechanisms of different natures. New relationships between these geometrical parameters and the scattering matrix elements are obtained. We also briefly reexamine the standard coherent polarimetric target decomposition algorithms in the light of quaternions. Next, an original use of polar decomposition for incoherent polarimetric imaging is proposed, which leads to the definition of the multilook boost parameter and of the degree of polarization dispersion. Subsequently, a new approach is presented, which consists in decomposing the scattering matrix into boost and rotation components before vectorization, then in averaging to generate boost and rotation coherency matrices separately. This leads to new inferred parameters: the boost and rotation entropies, and the concurrent dominant scattering mechanisms. The link between these new parameters and standard polarimetric invariants from the Cloude and Pottier decomposition is discussed. Eventually, the multilook extension of polar decomposition may allow this to be applied to the classification of remote sensing data. In this framework, a set of five parameters reducing to four in the monostatic case can be considered.