Single-look Synthetic Aperture Radar Images Research Articles

As If by Magic: Self-Supervised Training of Deep Despeckling Networks With MERLIN

Speckle fluctuations seriously limit the interpretability of synthetic aperture radar (SAR) images. Speckle reduction has thus been the subject of numerous works spanning at least four decades. Techniques based on deep neural networks have recently achieved a new level of performance in terms of SAR image restoration quality. Beyond the design of suitable network architectures or the selection of adequate loss functions, the construction of training sets is of uttermost importance. So far, most approaches have considered a supervised training strategy: the networks are trained to produce outputs as close as possible to speckle-free reference images. Speckle-free images are generally not available, which requires resorting to natural or optical images or the selection of stable areas in long time series to circumvent the lack of ground truth. Self-supervision, on the other hand, avoids the use of speckle-free images. We introduce a self-supervised strategy based on the separation of the real and imaginary parts of single-look complex SAR images, called MERLIN (coMplex sElf-supeRvised despeckLINg), and show that it offers a straightforward way to train all kinds of deep despeckling networks. Networks trained with MERLIN take into account the spatial correlations due to the SAR transfer function specific to a given sensor and imaging mode. By requiring only a single image, and possibly exploiting large archives, MERLIN opens the door to hassle-free as well as large-scale training of despeckling networks. The code of the trained models is made freely available at https://gitlab.telecom-paris.fr/RING/MERLIN.

Design of New Wavelet Packets Adapted to High-Resolution SAR Images With an Application to Target Detection

High resolution in synthetic aperture radar (SAR) leads to new physical characterizations of scatterers which are anisotropic and dispersive. These behaviors present an interesting source of diversity for target detection schemes. Unfortunately, such characteristics have been integrated and have been naturally lost in monovariate single-look SAR images. Modeling this behavior as nonstationarity, wavelet analysis has been successful in retrieving this information. However, the sharp-edge of the used wavelet functions introduces undesired high side-lobes for the strong scatterers present in the images. In this paper, a new family of parameterized wavelets, designed specifically to reduce those side lobes in the SAR image decomposition, is proposed. Target detection schemes are then explored using this spectro-angular diversity and it can be shown that in high-resolution SAR images, the non-Gaussian and robust framework leads to better results.

Doppler Spectrum-Based NRCS Estimation Method for Low-Scattering Areas in Ocean SAR Images

The image intensities of low-backscattering areas in synthetic aperture radar (SAR) images are often seriously contaminated by the system noise floor and azimuthal ambiguity signal from adjacent high-backscattering areas. Hence, the image intensity of low-backscattering areas does not correctly reflect the backscattering intensity, which causes confusion in subsequent image processing or interpretation. In this paper, a method is proposed to estimate the normalized radar cross-section (NRCS) of low-backscattering area by utilizing the differences between noise, azimuthal ambiguity, and signal in the Doppler frequency domain of single-look SAR images; the aim is to eliminate the effect of system noise and azimuthal ambiguity. Analysis shows that, for a spaceborne SAR with a noise equivalent sigma zero (NESZ) of −25 dB and a single-look pixel of 8 m × 5 m, the NRCS-estimation precision of this method can reach −38 dB at a resolution of 96 m × 100 m. Three examples are given to validate the advantages of this method in estimating the low NRCS and the filtering of the azimuthal ambiguity.

Snow cover area identification by using a change detection method applied to COSMO-SkyMed images

The information theoretic snow detection algorithm, a method that employs a change detection approach derived by Shannon’s information theory based on the conditional probability of the local means between two images taken at different times, is applied to multitemporal COSMO-SkyMed® data. The ultimate purpose of the method is the identification of snow cover areas in the case of extensive surface changes between summer and winter seasons. Both Himage and Ping Pong data in Stripmap acquisition mode from the COSMO-SkyMed constellation are processed. Results are compared to the available ground snow information gathered at the meteorological station present in the area. Quantitative assessments are obtained for Himage by considering a Landsat image as ground-truth. Receiver operating characteristic curves are used to deliver numerical comparisons between ground-truth and classified image, which is then compared to the well-known log-ratio approach. The proposed information theoretical approach to change detection provides very promising results in the case of large snow covering on multitemporal single-look synthetic aperture radar images at very high spatial resolution, due to its intrinsic low sensibility to speckle noise.

A Physical Approach for the Observation of Oil Spills in SAR Images

In this paper, a physical approach to support oil spills observation over synthetic aperture radar (SAR) images is presented. Electromagnetic model is based on an enhanced damping model that takes into account oil viscoelastic properties and wind speed. As a matter of fact, a multisensor approach is considered and a constant false alarm rate (CFAR) filter is used to minimize speckle effect. A set of experiments is presented and discussed. They show that oil spill processing is effective over single-look SAR images using mean input data.

λ-WMAP: a statistical speckle filter operating in the wavelet domain

In this paper we introduce the Γ-WMAP filter, a wavelet based equivalent to the classical Γ-MAP filter. We model speckle as additive signal-dependent noise, and propose to use the normal inverse Gaussian (NIG) distribution as a statistical model for the wavelet coefficients of both the reflectance image and the noise image. A method for estimating the parameters of the proposed statistical models is presented, and we show that the NIG distribution makes excellent fits to the distributions of the wavelet coefficients of single-look synthetic aperture radar (SAR) images. The performance of the Γ-WMAP filter is tested on three single-look SAR images. We find that when the filter is used in a global mode it may severely blur the image. However, when applied in a local, adaptive mode the new algorithm has excellent de-speckling performance. Visual comparisons with the Γ-MAP filter show that Γ-WMAP tends to give better de-speckling. Quantitative comparisons in homogeneous regions using both the equivalent number of looks and the log standard deviation as measures definitely show that the Γ-WMAP gives better speckle filtering.

Detection and imaging of arbitrarily moving targets with single-channel SAR

In synthetic aperture radar (SAR) images, moving targets cause imaging errors such as defocusing and/or displacement, depending on the direction of their motion. To obtain a correct and focused image of the moving targets, knowledge of the target parameters, position and velocity, is necessary. An algorithm for detection, parameter estimation, and imaging of ground-based targets, moving in arbitrary directions, is presented. The algorithm is based on evaluation of a sequence of single-look SAR images which is generated from conventional, single-channel SAR raw data. The position and velocity of the moving targets are estimated by applying two observation models, whereby aliasing due to a motion induced Doppler shift is considered. In this way, the moving targets can be detected almost independently of their moving direction. The estimated parameters are used to compensate for the imaging errors in the SAR images. Finally, the real motion of the targets in the scene is visualised in a compensated image sequence.