Deep learning approaches leverage data to learn complex relationships, achieving high predictive accuracy at the expense of explainability and transparency. In this paper, we present an analysis of a Transformer-based multi-target multi-sensor tracking method, the Transformer Tracker, utilizing the attention mechanism for explainability purposes. Our approach adapts the method to handle 3D space and integrates multiple sensors. We subsequently investigate the processes using attention weights through the attention rollout method. We perform this analysis across various scenarios, highlighting which measurements the Transformer Tracker focus on to predict the states of each object in those scenarios. The results align with data association and Kalman filtering processes, showcasing similar behaviors. This insight demonstrates the explainability capabilities of such approach, illustrating mechanisms similar to those of established and trusted tracking methods. The findings emphasize the reliability and transparency of Transformer-based tracking methods, bridging the gap between deep learning techniques and conventional tracking methods. This supports their application in complex and critical environments where understanding the decision-making process is crucial.

The motivation of this paper is to address the problem of registering airborne LiDAR data and optical aerial or satellite imagery acquired from different platforms, at different times, with different points of view and levels of detail. In this paper, we present a robust registration method based on building regions, which are extracted from optical images using mean shift segmentation, and from LiDAR data using a 3D point cloud filtering process. The matching of the extracted building segments is then carried out using Graph Transformation Matching (GTM) which allows to determine a common pattern of relative positions of segment centers. Thanks to this registration, the relative shifts between the data sets are significantly reduced, which enables a subsequent fine registration and a resulting high-quality data fusion.

This paper presents a performance evaluation of stock picking by merging several technical indicators. Several fusion operators have been proposed either in the probabilistic or in the possibilistic framework. The latter fuzzy framework has been introduced to manage the uncertain information embedded in financial time series due to human biases as studied by behavioral finance. Performances of portfolio resulting from the proposed systems are evaluated according to cumulative returns but also through a risk analysis point of view (Sharpe ratio). Two fusion mechanisms (one probabilistic, one possibilistic) aiming at discriminating common information from merged technical indicators, produce the higher portfolio performances. It also appears that selecting specific technical indicators affects the overall performances of the proposed stock picking systems. Indeed, studying the technical indicators selection through a shared/nonshared information point of view reveals that possibilistic framework is more robust to redundant sources than probabilistic framework. Effects of some parameters used in the fusion algorithms (amount of assets, window length analysis, ...) are also investigated. Results from all these tests clearly show the high potentiality of technical indicators fusion to improve portfolio performances. However, these first promising results have to be further inspected within wider contexts, as discussed at the end of the paper.

In this paper, we propose a novel model-based approach for 3D underwater scene reconstruction, i.e., bathymetry, for side scan sonar arrays in complex and highly reverberating environments like shallow water areas. The presence of multipath echoes and volume reverberation generates false depth estimates. To improve the resulting bathymetry, this paper proposes and develops an adaptive filter, based on several original geometrical models. This multimodel approach makes it possible to track and separate the direction of arrival trajectories of multiple echoes impinging the array. Echo tracking is perceived as a model-based processing stage, incorporating prior information on the temporal evolution of echoes in order to reject cluttered observations generated by interfering echoes. The results of the proposed filter on simulated and real sonar data showcase the clutter-free and regularized bathymetric reconstruction. Model validation is carried out with goodness of fit tests, and demonstrates the importance of model-based processing for bathymetry reconstruction.

In this paper a 3-D reconstruction method is proposed of sea bottom topography, i.e. bathymetry, for sonar data in highly reverberant and multi-path underwater environments. Recent publications showcase the negative impact of waves involving sea surface reflections on the sea bottom imaging process. The novelty of our proposal is twofold: firstly it relies on the use of Markovian-model based processors for bahymetric reconstruction and involves data association filters. Secondly, we propose a nearest neighbor version of the Integrated Probabilistic Data Association filter, capable of filtering several echo trajectories in the presence of clutter with a relatively reduced complexity compared to other existent methods.

ONERA radar RAMSES was recently upgraded with low frequency band (P-band, 435 MHz). In P-band, frequencies penetrate through the ground and through forest canopy. Unfortunately, the formation of the subsurface radar images presents a number of new specific challenges that include algorithm validity, calibration methods, radio-frequency interference, and image focusing and analysis. This paper addresses the aspects of focusing images from an airborne low frequency SAR of buried targets in a lossy and dispersive soil. The hydrological model (Richards' equation) was used to model the multilayer case.

In this paper, we propose a method for estimating the nonlinear differential equation governing the propagation of internal waves (IWs) in a stratified water column of the ocean. This estimation of the kind of nonlinear differential equation and the coefficients of this equation is performed from profiles of the synthetic aperture radar signature at the ocean surface. Theoretically, the equation is based upon the Korteveg-de Vries (KdV) equation or one of its improved versions. The estimation of the differential equation version is based on nonlinear polynomial autoregressive models up to the third order. The estimation of the parameters of these models is performed simply by solving a set of linear equations using higher order statistics. This general approach allows us to identify the best model of propagation, i.e., the kind of nonlinearity as well as the structure of each kernel (linear, quadratic, cubic), without determining an analytic solution to the equation of propagation. Moreover, the physical oceanic parameters (such as the thermocline depth) are deduced from the estimated coefficients of the KdV equation (assuming an underlying model of the water column). Results on simulated profiles generally lead to an exact model identification (i.e., the one used for simulation with the exact order of nonlinearity and the exact structure of each kernel) and lead to satisfying geophysical parameter estimates. Finally, the geophysical parameters estimated from three sets of ERS-1 profiles are generally coherent with the parameters observed in the IW propagation mechanism but have not been validated by "in situ" measurements.

In the framework of the SAR imaging process simulator we are developing, the modification of the short wave spectrum by ocean surface is of great importance. Short waves at the sea surface produce the backscattering of the radar wave. So the modulation of the short wave distribution by ocean surface induces the modulation of the SAR image. In this paper we present a two-step algorithm designed to simulate the modulation of the short waves by both current and long waves. First of all the modulation of the long waves by current is taken into account. Then the modification of the short wave spectrum by long-wave orbital velocity and current is considered. To achieve the calculation we use the action balance equation (ABE) based on a weak interaction theory. ABE theory and the two-step algorithm are described here

ONERA, the French aerospace research agency, has developed a multi-frequency, full polarimetric and high resolution airborne SAR facility named RAMSES, which has been expanded with a P-band (435 MHz) capability at the end of year 2000. The Gironde area was chosen as a suitable test-site for fieldwork validation of P-band capabilities in coordination with the "low frequency radar working group" set up by CNES (French Space Agency), the DGA (DOD Procurement Agency) and the CEV (French Flight Test Center). This experiment has explored the capabilities of this low frequency band for sub-surface moisture detection (on the "Pyla" sand dune), biomass assessment (on the Nezer forest), mapping of the ocean bathymetry/salinity (on the "bassin d'Arcachon", estuary of the Gironde), and archaeology (imaging of buried Roman-era artifacts). After a short description of the radar P and of the experience feedback of campaign "PYLA 2001" on the P-band radar, we present a preliminary synthesis of the results on the different sites. The "PYLA 2001" campaign results are encouraging. We therefore plan further experiments at the beginning of 2004 over Egypt (with simultaneous band P+L). In conclusion, we present the future research trend at ONERA in the domain of lower frequency (VHF/UHF) SAR.

Microwave remote sensing at very long wavelength is already used and validated for coastal zone regions and regions covered by ice (Antarctica) using airborne systems. For many other applications, a high potential exists and the PYLA 2001 experiment allowed the acquisition of data for fields still under preliminary tests. The experiment was carried on a Transall aircraft (French army/DGA) and used the SAR sensor RAMSES (multifrequency/multipolarization) from ONERA by P. Paillou et al. (2001). For the experiment, the acquisition was performed in P band (435 MHz). The aim was to explore potentials of the low frequency domain for sub-surface moisture detection (Pyla sand dune), biomass evaluation (Nezer forest), mapping of the ocean bathymetry and salinity (Arcachon inlet, estuary of the river Gironde), and archaeology (St Germain d'Esteuil, Dignac, Moulin du Fa). A dedicated P-band calibration site was set up in order to fully exploit the polarimetric information. This part of the study consists of analyzing SAR images from RAMSES (P band) in order to be able to take into account the effect of longer surface waves, winds and currents on the surface roughness in the Arcachon inlet. Tidal currents associated with long oceanic waves recorded at the mouth of the inlet, including the northern channel and sand banks of the ebb tidal delta. The analysis provides tools allowing the production of current maps, which are in turn compared to in-situ measurements by GPS on surface drifting buoys and to current velocity fields provided by a 2DH numerical model, MARS-2D. The evaluation of the ability of a 1D hydrodynamical model applied to SAR images in order to reproduce currents in the area presented. Multipolarization images can be used in order to separate the information associated with the ocean surface (tilt) modulation, which is different for each polarimetric image, from the hydrodynamic modulation, similar for each image. The tilt modulation is representative of a Real Aperture Radar-like (RAR) image. The data being available quite late, we just performed correlations between each set of data by using polarimetric coefficient. We obtain good answers on the information associated with the motion of the scatterers.

A Markov chain model is applied for the segmentation of oil slicks acquired by SAR sensors. Actually, oil slicks have specific impact on ocean wave spectra. Initial wave spectra may be characterized by three kinds of waves, big, medium and small, which correspond physically to gravity and gravity-capillary waves. The increase of viscosity due to the presence of oil damps gravity-capillary waves. This induces a damping of the backscattering to the sensor, but also a dampening of the energy of the wave spectra. Thus, local segmentation of wave spectra may be achieved by the segmentation of a multiscale decomposition of the original SAR image. In this work, the unsupervised segmentation is achieved by using a vectorial extension of the Hidden Markov Chain (HMC) model. Parameters estimation is performed using the general Iterative Conditional Estimation (ICE) method. The problem of estimating multi-dimensional and non-Gaussian densities is solved by using a Principal Component Analysis (PCA). The algorithm has been applied on an ERS-PRI image. It yields interesting segmentation results with a very limited number of false alarms. Also, the multiscale segmentation proved to be an interesting alternative to classify marginal or degraded slicks.

International audience

Higher order moments have been, for the last decade, an important field of interest, but generally limited to one dimensional signal cases. We introduce in this paper a 2D bispectrum to detect nonlinearity in the SAR image mapping process, by using the bicoherency of a 2D signal which is theoretically flat over all frequencies, if the process is linear. Two bicoherency estimators are developed, the first one using a direct method for bispectral estimation and a periodogram for the spectral estimator, and the second one based on an indirect method and correlogram. In order to validate our nonlinearity detection method, we have tested it first on two simulated images, a linear one and a nonlinear one, and then on two SAR images, one seeming to be nonlinear. Conclusions are drawn by comparing the SAR image and the simulated image bicoherencies.