Multi-agent trajectory prediction is a key task in computer vision for autonomous systems, particularly in dense and interactive environments. Existing methods often struggle to jointly model goal-driven behavior and complex social dynamics, which leads to unrealistic predictions. In this paper, we introduce VISTA, a recursive goal-conditioned transformer architecture that features (1) a cross-attention fusion mechanism to integrate long-term goals with past trajectories, (2) a social-token attention module enabling fine-grained interaction modeling across agents, and (3) pairwise attention maps to show social influence patterns during inference. Our model enhances the single-agent goal-conditioned approach into a cohesive multi-agent forecasting framework. In addition to the standard evaluation metrics, we also consider trajectory collision rates, which capture the realism of the joint predictions. Evaluated on the high-density MADRAS benchmark and on SDD, VISTA achieves state-of-the-art accuracy with improved interaction modeling. On MADRAS, our approach reduces the average collision rate of strong baselines from 2.14% to 0.03%, and on SDD, it achieves a 0% collision rate while outperforming SOTA models in terms of ADE/FDE and minFDE. These results highlight the model’s ability to generate socially compliant, goal-aware, and interpretable trajectory predictions, making it well-suited for deployment in safety-critical autonomous systems.

Self-supervised learning (SSL) has shown great promise in computer vision, enabling networks to learn meaningful representations from large unlabeled datasets. SSL methods fall into two main categories: instance discrimination and image modeling. While instance discrimination is fundamental to SSL, it was originally designed for classification and may be less effective for downstream tasks that require fine-grained or spatially localized representations. In this focused survey, we study SSL for object detection under challenging practical conditions, with particular emphasis on small object detection, domain shift and few-shot learning. Building upon previous surveys, we not only provide a detailed comparison of SSL strategies, but also assess their effectiveness for object detection using both CNN and ViT-based architectures. Our benchmark is performed fairly by fine-tuning a Faster R-CNN initialized with several exemplary SSL methods ourselves, including object-level Instance Discrimination and Masked Image Modeling methods, on the widely used COCO dataset, as well as on a domain-specific dataset focused on vehicle detection in infrared remote sensing imagery. We also evaluate the impact of pre-training on custom domain-specific datasets, highlighting how some SSL strategies are better suited for handling uncurated data. Furthermore, we assess the methods in few-shot settings and inference on noisy input, revealing important behavioral differences depending on the type of encoder used. Our findings highlight that combining approaches with complementary local and global biases improves performance across the evaluated object detection settings. Overall, this survey provides a practical guide for selecting optimal SSL strategies in different scenarios. • We propose a survey on self-supervised learning for real-world object detection. • In our benchmarks, we pay attention to small object detection performance. • Challenging conditions such as frugal setting or remote sensing data are considered. • The benefits of pre-training on custom domain-specific datasets is assessed. • A road map for selecting appropriate self-supervised learning strategies is provided.

Infrared Small Target Detection (IRSTD) is a challenging task in defense applications, where complex backgrounds and tiny target sizes often result in numerous false alarms using conventional object detectors. To overcome this limitation, we propose an Anomaly-Aware version of You Only Look Once (YOLO) detector (namedAA-YOLO), which integrates a statistical anomaly detection test into its detection head. By treating small targets as unexpected patterns against the background, AA-YOLO effectively controls the false alarm rate. Our approach not only achieves competitive performance on several IRSTD benchmarks, but also demonstrates remarkable robustness in scenarios with limited training data, noise, and domain shifts. Furthermore, since only the detection head is modified, our design is highly generic and has been successfully applied across various YOLO backbones, including lightweight models. It also provides promising results when integrated into an instance segmentation YOLO. This versatility makes AA-YOLO an attractive solution for real-world deployments where resources are constrained. The code is available at https://github.com/AMIAD-Research/AA-YOLO . • We present AA-YOLO, a YOLO model with anomaly testing for IR small target detection. • AA-YOLO attains state-of-the-art detection of small targets in complex IR scenes. • Our method is robust across varied scenarios: limited data, noise, and domain shifts. • AA-YOLO is lightweight and works with many YOLO backbones, ideal for real-world use.

Deep neural networks perform well in many applications but often fail when exposed to out-of-distribution (OoD) inputs. We identify a geometric phenomenon in the embedding space: in-distribution (ID) data show higher variance than OoD data under stochastic perturbations. Using high-dimensional geometry and statistics, we explain this behavior and demonstrate its application in improving OoD detection. Unlike traditional post-hoc methods, our approach integrates uncertainty-aware tools, such as Bayesian approximations, directly into the detection process. Then, we show how considering the unit hypersphere enhances the separation of ID and OoD samples. Our mathematically sound method achieves competitive performance while remaining simple.

Identifying pigments in Cultural Heritage artifacts is key to uncovering their origin and guiding conservation strategies. Although recent advances in non-invasive imaging have enabled the collection of rich multimodal data, existing methods often fall short in dealing with uncertain, ambiguous, or noisy information. This paper introduces a versatile fusion framework grounded in Belief Function Theory, combining domain-informed evidence modeling with neural optimization. Specifically, we propose a general strategy for assigning mass functions by leveraging expert knowledge encoded in parametric Evidence Mapping Functions, which are further refined through task-specific training using constrained neural networks. When applied to pigment classification, our method demonstrates robustness against source variability and class ambiguity. Experiments conducted on both synthetic and mock-up datasets validate its effectiveness and suggest promising potential for broader applications.

Abstract Puzzle-solving is a problem having applications for instance in archaeology and cultural heritage. Proposed solutions often suffer from a performance loss when the pieces are eroded, a characteristic that is pervasive across various use—cases such as frescoes reconstruction. Most approaches divide the problem into two fundamental phases: discriminating and then positioning the pieces. We focus on the case of puzzles with square pieces, without any missing or extraneous pieces, and we introduce the first two-step deep learning solution capable of efficiently solving puzzles, from discrimination to piece placement, while remaining robust to erosion. In the context of permutation learning, we propose to use transformers to determine the correct placement of the pieces and an encoder that uses the information at the edge of the pieces. This method sets a new state of the art, achieving a significant performance gain, and introduces a new approach for learning similarity functions in the context of puzzle solving.

Trajectory prediction is a critical task for autonomous systems such as self-driving cars, surveillance systems, and social robots. The goal is to predict the future paths of road users, including cars, bikes, and pedestrians, by using their historical movement patterns and the surrounding environment. While traditional models based on Newton’s laws and social interaction forces have been used in the past, data-driven methods have become essential for learning complex spatial and temporal interactions between pedestrians. In this work, we propose SwYn-Net, a model for predicting long-term trajectories of pedestrians that can handle the uncertainty of multiple plausible paths and address the accumulation of errors over time. Our approach uses a shifted-window attention mechanism to capture the scene’s local and global context. We evaluate our model on the SDD, inD and the MOT20 datasets and show that SwYn-Net can be an efficient model to predict short-term and long-term trajectories.

Detecting small to tiny targets in infrared images is a challenging task in computer vision, especially when it comes to differentiating these targets from noisy or textured backgrounds. Traditional object detection methods such as YOLO struggle to detect tiny objects compared to segmentation neural networks, resulting in weaker performance when detecting small targets. To reduce the number of false alarms while maintaining a high detection rate, we introduce an a contrario decision criterion into the training of a YOLO detector. The latter takes advantage of the unexpectedness of small targets to discriminate them from complex backgrounds. Adding this statistical criterion to a YOLOv7-tiny bridges the performance gap between state-of-the-art segmentation methods for infrared small target detection and object detection networks. It also significantly increases the robustness of YOLO towards few-shot settings.

The detection of tiny objects is a challenging task in computer vision. Conventional object detection methods have difficulties in finding the balance between high detection rate and low false alarm rate. In the literature, some methods have addressed this issue by enhancing the feature map responses for small objects, but without guaranteeing robustness with respect to the number of false alarms induced by background elements. To tackle this problem, we introduce an a contrario decision criterion into the learning process to take into account the unexpectedness of tiny objects. This statistic criterion enhances the feature map responses while controlling the number of false alarms (NFA) and can be integrated as an add-on into any semantic segmentation neural network. Our add-on NFA module not only allows us to obtain competitive results for small target, road crack and ship detection tasks respectively, but also leads to more robust and interpretable results.

This work deals with epipole estimation related to egocentric camera localization in surveillance and security applications. Matching visual features in the images provides some evidences for various solutions, so that epipole localization can be addressed as a fusion task with a large number of sources including outlier ones. In order to deal with source imprecision and uncertainty, we rely on the belief function theory and a 2D framework suited for our application. In this framework, we address the challenges introduced by a large number of sources with a strategy based on clustering and intra-cluster fusion. The proposed method exhibits more robustness in terms of accuracy and precision when compared on real data with the standard algorithms which provide single solution. Since we provide a Basic Belief Assignment as a result, our strategy is particularly adapted for the prospective combination with additional sources of information.

Preventive conservation is an important practice in Cultural Heritage. The constant monitoring of the state of conservation of an artwork helps us reduce the risk of damage and number of necessary interventions. In this work, we propose a probabilistic approach for the detection of alterations on the surface of historical violins based on an a-contrario framework. Our method is a one step NFA clustering solution which considers grey-level and spatial density information in one background model. The proposed method is robust to noise and avoids parameter tuning and any assumption about the quantity of the worn-out areas. We have used as input UV induced fluorescence (UVIFL) images for considering details not perceivable with visible light. Tests were conducted on image sequences included in the “Violins UVIFL imagery” dataset. Results illustrate the ability of the algorithm to distinguish the worn area from the surrounding regions. Comparisons with state-of-the-art clustering methods show improved overall precision and recall.

In urban cluttered scenes, a photo provided by a wearable camera may be used by a walking law-enforcement agent as an additional source of information for localizing themselves, or elements of interest related to public safety and security. In this work, we study the problem of locating the epipole, corresponding to the position of the moving camera, in the field of view of a reference camera. We show that the presence of outliers in the standard pipeline for camera relative pose estimation not only prevents the correct estimation of the epipole localization but also degrades the standard uncertainty propagation for the epipole position. We propose a robust method for constructing an epipole location map, and we evaluate its accuracy as well as its level of integrity with respect to standard approaches.

In this work, we use the Belief Function Theory which extends the probabilistic framework in order to provide uncertainty bounds to different categories of crowd density estimators. Our method allows us to compare the multi-scale performance of the estimators, and also to characterize their reliability for crowd monitoring applications requiring varying degrees of prudence.

Ahstract- This paper introduces an innovative approach for handling 2D compound hypotheses within the Belief Function Theory framework. We propose a polygon-based generic representation which relies on polygon clipping operators. This approach allows us to account in the computational cost for the precision of the representation independently of the cardinality of the discernment frame. For the BBA combination and decision making, we propose efficient algorithms which rely on hashes for fast lookup, and on a topological ordering of the focal elements within a directed acyclic graph encoding their interconnections. Additionally, an implementation of the functionalities proposed in this paper is provided as an open source library. Experimental results on a pedestrian localization problem are reported. The experiments show that the solution is accurate and that it fully benefits from the scalability of the 2D search space granularity provided by our representation.

Large scale social events are characterized by very high densities (at least locally) and an increased risk of congestions and fatal accidents. Our work focuses on the specific problem of pedestrian detection in high-density crowd images, denoted by strong homogeneity and clutter. We propose and compare different evidential fusion algorithms which are able to exploit multiple detectors based on different gradient, texture and orientation descriptors. The evidential framework allows us to model spatial imprecision arising from each of the detectors, both in the calibration and in the spatial domains. Moreover, we propose a Belief Function allocation that takes into account both types of imprecision. Results on difficult high-density crowd images acquired at Makkah during the Muslim pilgrimage show that the proposed combined fusion algorithm leads to better results than taking into account only individual sources of imprecision.

This paper addresses the problem of pedestrian detection in high-density crowd images, characterized by strong homogeneity and clutter. We propose an evidential fusion algorithm which is able to exploit multiple detectors based on different gradient, texture and orientation descriptors. The evidential framework allows us to model the spatial imprecision arising from each of the detectors. A first result of our study is that the fusion results underline clearly the good complementarity among the four descriptors we considered for this specific context. Moreover, the proposed algorithm outperforms a fusion solution based on Multiple Kernel Learning on difficult high-density crowd images acquired at Makkah at the height of the Muslim pilgrimage.

Efficient crowd counting is an essential task in crowd monitoring, and significant advances have been made in this field recently by counting-by-regression techniques. We propose in this work a learning-to-count strategy with a generic detection algorithm which benefits from a counting regressor in order to identify crowded subregions with inadequate head detection performance, and to improve their representativeness in the training set. A straightforward but crucial step is proposed in order to take into account perspective correction within the proposed framework. An evaluation on Makkah images with medium to very high densities demonstrates the effectiveness of our algorithm and its capability to reach a count error of less than 5% in this difficult setting.

Despite many proposed solutions, multi-object tracking remains a challenging problem in complex situations involving partial occlusions and non-uniform and abrupt illumination changes. Considering modular systems, the tracking performance strongly depends on the consistency of the different blocks relatively to error features. In this work, using the Belief Function framework, we take into account the reliability and the imprecision of the object detection and location to characterize objects and to derive a reliable descriptor. Since this latter is then estimated only on safe object subparts, even in case of crosses between objects, we use a distance between descriptor robust to partial occlusion, namely the recently proposed Bin-Ratio-Distance. Results obtained on various actual sequences underline the interest of the proposed algorithm by outperforming the tested alternative approaches.

Robust wide baseline pose estimation is an essential step in the deployment of smart camera networks. In this work, we highlight some current limitations of conventional strategies for relative pose estimation in difficult urban scenes. Then we propose a solution which relies on an adaptive search of corresponding interest points in synchronized video streams which allows us to converge robustly towards a high-quality solution. The experiments are performed using a manually annotated ground truth of a large scale scene exhibiting significant depth and perspective variation, uniform areas, repetitive patterns and homogeneous dynamic elements. The results show a fast and robust convergence of the solution, and a significant improvement, compared to single image based alternatives, of the RMSE of ground truth matches, and of the maximum absolute error.

This paper studies the problem of crack detection in images characterized by high gradient backgrounds. We propose an extension of a Marked Point Process model which has been successfully used for wrinkle detection. We show that our method exhibits state of the art results on a difficult image dataset, by proposing a robust trade-off between local analysis approaches, which exploit a limited amount of information around the area of interest, and global reconnection strategies, which aim to detect the crack at image level. Additional tests on a standard dataset show that the proposed method exhibits excellent performance on images with a more uniform background as well, underlining its usefulness in varying contexts.

In Global Navigation Satellite System (GNSS) positioning, urban environments represent an issue particularly because of multipath and nonline-of-sight effects. The latter effects induce erroneous pseudorange observations that then should be discarded in order not to affect the estimation of the receiver position. This paper proposes a new approach for the detection of outliers in the pseudorange observations. Based on two models representing the distribution of inconsistent data (naive models), two criteria are proposed to partition the data between inliers and outliers and to estimate the location parameters. These criteria are then implemented in two localization algorithms. In addition, by considering hypotheses specific to GNSS localization, pseudorange selection and a regularization step are implemented in order to reduce the complexity and to improve the problem conditioning. Using simulated and actual datasets, the proposed algorithms are compared with popular and recent methods addressing the GNSS positioning problem. We show that the outlier detection improves the estimation of the receiver location and outperforms the classical approaches particularly when the environment is constrained.

This study focuses on the problem of object reconstruction through several frames of a video sequence. Elementary detections on which this reconstruction is based are assumed to be fragments of the objects. Belief function framework allows then the modelling of the uncertain and imprecise location of these object fragments within the image. We show that the two competing mechanisms for object reconstruction, namely the data accumulation and their temporal removal or weighting, can be implemented using belief function operators. Results illustrate the robustness of the proposed approach to object partial occultation and crossing.

The spatial resolution of thermal infrared (TIR) instruments is often not sufficient for many applications, but this low resolution is counterbalanced by the high temporal resolution (for example the SEVIRI instrument onboard the European Meteosat 8 and 9 presents a spatial resolution of 3 km$ \times$3 km at nadir and a temporal resolution of 15 mn). At kilometric scales, the observed pixel is generally heterogeneous in terms of land cover, and the temperatures of the different components may present large discrepancies. This paper presents a methodology to infer the temperatures of the various land cover/use classes composing a mixed pixel, from a whole pixel measurement. To infer intra-pixel temperature, information on the mixture within each low resolution pixel, e.g., the proportions of the land cover types derived from high spatial resolution imaging, account for a first constraint. However, in the absence of supplementary constraints, the number of unknown variables is greater than the number of measurements, and there is not uniqueness of the solution. Thus, we propose to take advantage of a priori knowledge provided by a land surface model (LSM), and of the temporal and spatial correlation features of the surface temperature. We propose a new downscaling method for estimating sub pixel signal. It applies to TIR data and: the inversion procedure provides as a result, the land surface temperature (LST) temporal series of each land cover/use class (called endmember) constituting the coarse resolution pixel. Three kinds of a priori information have been introduced, namely 1) a first guess subpixel temperature derived from the SEtHyS LSM; 2) a Markov Random Chain model of the surface temperature temporal dependencies from times$t$to$t + 1$; 3) a Markov Random Field model of the spatial dependencies between endmember temperatures. Then, the “Maximum A Posteriori” estimator provides the most likely endmember temperatures, given 1) the observed coarse resolution temperatures, 2) the composition of the pixels in terms of “land cover/land use,” and 3) the LSM first guess subpixel temperature values, 4) the a priori spatial and temporal Markov models. The performance of this new method has been first evaluated on simulated data (random Gaussian variables with means equal to endmember temperatures simulated using LSM). The method accuracy versus the observation errors and the number of endmembers was analyzed. The algorithm was then run on actual data, namely Meteosat SEVIRI Land Surface products acquired over an agricultural region in southeastern France. The performance evaluation was done by comparing the subpixel LST estimations to the high-resolution temperatures provided by the Terra/ASTER instrument. Due to the huge bias between sensors ($ \sim$4 K), an intercalibration preprocessing between SEVIRI and ASTER was done. In this case, the achieved RMSE is lower than 2 K.

This paper presents a new method for unsupervised subpixel change detection using image series. The method is based on the definition of a probabilistic criterion capable of assessing the level of coherence of an image series relative to a reference classification with a finer resolution. In opposition to approaches based on an a priori model of the data, the model developed here is based on the rejection of a nonstructured model-called a-contrario model-by the observation of structured data. This coherence measure is the core of a stochastic algorithm which automatically selects the image subdomain representing the most likely changes. A theoretical analysis of this model is led to predict its performances, in particular regarding the contrast level of the image as well as the number of change pixels in the image. Numerical simulations are also presented that confirm the high robustness of the method and its capacity to detect changes impacting more than 25 percent of a considered pixel under average conditions. An application to land-cover change detection is then provided using time series of satellite images.

The paper presents a new methodology adapted to the downscaling of low resolution IRT signals, i.e. the estimation of subpixel temperatures. The approach is based on the inversion of subpixel variables by multilinear regressions constrained by a priori temperature estimates provided by a physical land surface model. The method was developed and validated against a synthetic database built on model simulations. The precision of the methodology was analysed in terms of errors on the subpixel temperature estimations according to model and observation uncertainties. The impact of the number of observations used (i.e. the number of low resolution pixels considered) as well as the influence of the pixel heterogeneity were studied.

The goal of this paper is to propose a methodology based on vegetation index fusion to provide an accurate estimation of the fraction of vegetation cover (fCover). Because of the partial and imprecise nature of remote-sensing data, we opt for the evidential framework that allows us to handle such kind of information. The defined fCover belief functions are continuous with the interval [0, 1] as a discernment space. Since the vegetation indices are not independent (e.g., perpendicular vegetation index and weighted difference vegetation index are linearly linked), we define a new combination rule called “cautious adaptive” to handle the partial “nondistinctness” between the sources (vegetation indices). In this rule, the “nondistinctness” is modeled by a factor$\varrho$varying from zero (distinct sources) to one (totally correlated sources), and the fusion rule varies accordingly from the conjunctive rule to the cautious one. In terms of results, both in the cases of simulated data and actual data, we show the interest of the combination of two or three vegetation indices to improve either the accuracy of fCover estimation or its robustness.

In this paper, a new method is presented for a subpixelic land cover classification using both high-resolution structural information and coarse-resolution (CR) temporal information. To that aim, the linear mixture model is used for pixel disaggregation. It enables us to describe a CR time series in terms of the mixture of classes that are represented within each pixel. Then, the Bayes' rule and the maximuma posterioricriterion lead to the definition of an energy function whose minimum corresponds to the researched optimal classification. A theoretical analysis of the labeling errors that may be obtained using this energy function is provided, raising the main parameters for labeling performance. The optimal classification is computed by combining linear regressions and simulated annealing, leading to an unsupervised algorithm. The method is validated with numerical results obtained on two different agricultural scenes (i.e., the Danubian plain and the Coet Dan watershed).

The SAIL model (proposed by Verhoef) is largely used in the remote sensing community to calculate the canopy Bidirectional Reflectance Distribution Function. The simulation results appear acceptable compared to observations especially for not very dense planophile vegetation. However, for erectophile dense crops (e.g. corn) the simulations appear less accurate. This inadequacy is due to the assumption that the multiple scattered are isotropically distributed. The SAIL parameters are interpretable at the level of elementary layer components. Now, the Adding method (initially proposed by Van de Hulst) provides a good framework to model the radiative transfer inside a vegetation layer, but its parameter estimation lies on very simple geometric modeling of the canopy. In this paper, we propose an adaptation of the Adding method using the SAIL model canopy representation. Such an approach allows both to overcome the isotropy assumption and to take into account the multi hot spot effect. It also allows to check the energy conservation in both turbid and discrete case.

Ant colony optimization (ACO) has been proposed as a promising tool for regularization in image classification. The algorithm is applied here in a different way than the classical transposition of the graph color affectation problem. The ants collect information through the image, from one pixel to the others. The choice of the path is a function of the pixel label, favoring paths within the same image segment. We show that this corresponds to an automatic adaptation of the neighborhood to the segment form, and that it outperforms the fixed-form neighborhood used in classical Markov random field regularization techniques. The performance of this new approach is illustrated on a simulated image and on actual remote sensing images.

Digital change detection deals with the quantification, from multi-date imagery, of temporal phenomena, such as Aforestation-Reforestation-Deforestation, agricultural field rotation, abnormal evolution of the land surface. Despites the numerous change detection indices already proposed, none is sufficiently precise and reliable. We propose then to detect changes by considering not only one but several change indices, as well as information available from other source than remote sensing, i.e. derived from surface evolution model or a priori. For fusion, we chose the framework of the Dempster-Shafer evidence theory. It allows for some global ignorance, which is either present at the borders between the 'No-Change' and 'Change' classes, or is due to the poor quality of some change indices. The performance of the Non Remote Sensing (NRS) data change prediction, when known (e.g. statistical error of a model), can be taken into account the discounting of the mass functions. We present the results obtained in two different cases of application: forest logging and winter vegetation cover of fields in intensive farming areas. Remote sensing data are SPOT/HRV images. Considering the performance in terms of Non-Detection and False Detection rates, the interest of combining at least two change indices was clearly stated. The interest of NRS information has been then evaluated in the case of the field winter coverage application.

P-band airborne SAR (the French RAMSES facility) data were acquired in the framework of the PYLA 2001 project. Images from the Nezer forest are investigated by means of a Van Zyl classification scheme. A new relation between the classification in terms of backscattering mechanisms and the mean height of the trees is proposed.

This paper discusses the potential of radar signal to characterise the bare surface roughness in arid or semi-arid regions. The used microwave sensor is the SAR of ERS. Ground truth measurements were acquired over different arid sites in the South of Tunisia. An empirical approach is proposed to derive the surface roughness from SAR measurements. The relationships with two different kinds of roughness have been studied: the geometric roughness, which is characterised by a rather new parameter called Zs, and the classical aerodynamic roughness Z/sub 0/.

This paper presents an original methodology to retrieve surface (< 5 cm) soil moisture over low vegetated regions using the two active microwave instruments of ERS satellite. The developed algorithm takes advantage of the multi-angular configuration and high temporal resolution of the Wind Scatterometer (WSC) combined with the SAR high spatial resolution. High correlations (R/sup 2/ greater than 0.8) are observed for three studied watersheds in France with an rms error smaller than 4% between real and retrieved moistures.

In the remote sensing domain, the combination of multi-scale satellite data appears as a new challenge. It should provide significant improvements in Earth monitoring by use of the complementary of the data presenting either high spatial resolution or high time repetitiveness. Here, we propose an algorithm based on the Dempster-Shafer evidence theory, which allows the modeling of the mixed feature of the low spatial resolution pixels, and the modeling of the class confusion when time information is not sufficient, by consideration of compound hypotheses such as unions of classes. It has been applied on SPOT/HRV image and NOAA/AVHRR series, and the results have clearly shown the improvement brought by the proposed data fusion.

Previous studies have shown the possibility of using European Remote Sensing/synthetic aperture radar (ERS/SAR) data to monitor surface soil moisture from space. The linear relationships between soil moisture and the SAR signal have been derived empirically and, thus, were a priori specific to the considered watershed. In order to overcome this limit, this study focused on two objectives. The first one was to validate over two years of data the empirical sensitivity of the radar signal to soil moisture, in the case of three agricultural watersheds with different soil compositions and land cover uses. The slope of the observed relationship was very consistent. Conversely, the offset could change, making the soil moisture retrieval only relative (and not absolute). The second one was to propose an "operational" methodology for soil moisture monitoring based on ERS/SAR data. The implementation of this methodology is based on two steps: the calibration period and the operational period. During the calibration period, ground truth campaigns are performed to measure vegetation parameters (to correct the SAR signal from the vegetation effect), and the ERS/SAR data is processed only once a field land cover map is established. In contrast, during the operational period, no vegetation field campaigns are performed, and the images are processed as soon as they are available. The results confirm the relevance of this operational methodology, since no loss of performance (in soil moisture retrieval) is observed between the calibration and operational periods.

The aim of this paper is to show that Dempster-Shafer evidence theory may be successfully applied to unsupervised classification in multisource remote sensing. Dempster-Shafer formulation allows for consideration of unions of classes, and to represent both imprecision and uncertainty, through the definition of belief and plausibility functions. These two functions, derived from mass function, are generally chosen in a supervised way. In this paper, the authors describe an unsupervised method, based on the comparison of monosource classification results, to select the classes necessary for Dempster-Shafer evidence combination and to define their mass functions. Data fusion is then performed, discarding invalid clusters (e.g. corresponding to conflicting information) thank to an iterative process. Unsupervised multisource classification algorithm is applied to MAC-Europe'91 multisensor airborne campaign data collected over the Orgeval French site. Classification results using different combinations of sensors (TMS and AirSAR) or wavelengths (L- and C-bands) are compared. Performance of data fusion is evaluated in terms of identification of land cover types. The best results are obtained when all three data sets are used. Furthermore, some other combinations of data are tried, and their ability to discriminate between the different land cover types is quantified.