Publications
Journal articles
2024
- JSTARSWet Snow Detection From Satellite SAR Images by Machine Learning With Physical Snowpack Model LabelingGallet, Matthieu, Atto, Abdourrahmane, Karbou, Fatima and 1 more authorIEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 2024
The detection of wet snow by satellite imaging is currently done in an unsupervised way and lacks quantitative evaluation due to the difficulty of collecting ground truths in extreme environments. In this article, we propose to take into account information associated with a physical model to label satellite data for the purpose of supervised learning of snow properties using synthetic aperture radar (SAR) imagery. This dataset is constructed from Sentinel-1 SAR images, augmented with topographic information obtained from a digital elevation model. The labeling of this data is done at the scale of the Northern Alps using the CROCUS physical snow model. Then, we trained, over 13 combinations of labeled dataset, a wide range of machine learning models to quantitatively identify the most relevant learners for the wet snow detection task. The results demonstrate consistency among the different algorithms, with significant improvement observed when incorporating polarimetric combinations and topographic orientation data in the input of the model. The best algorithmic solution trained on this dataset is evaluated by comparing the obtained wet snow map over a validation area in the French massif of the Grandes Rousses with the existing Copernicus products, fractional snow cover, and SAR wet snow. We also compare the temporal results obtained at one meteorological station located in the test area. The results show a better representation of wet snow during the melting period using the supervised learning approach, as well as a reduction in areas classified as wet during the winter season.
@article{10360230, title = {Wet Snow Detection From Satellite SAR Images by Machine Learning With Physical Snowpack Model Labeling}, author = {Gallet, Matthieu and Atto, Abdourrahmane and Karbou, Fatima and Trouvé, Emmanuel}, journal = {IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing}, year = {2024}, doi = {10.1109/JSTARS.2023.3342990}, keywords = {Snow,Labeling,Satellites,Radar polarimetry,Machine learning,Liquids,Backscatter}, url = {}, volume = {17}, number = {}, pages = {2901-2917}, month = {}, issn = {2151-1535} }
2023
- TGRSNew Robust Sparse Convolutional Coding Inversion Algorithm for Ground Penetrating Radar ImagesIEEE Transactions on Geoscience and Remote Sensing 2023
In this paper, we propose two algorithms to enhance the interpretability of the hyperbola in B-scans obtained with a Ground Penetrating Radar (GPR). These hyperbolas are the responses of buried objects or cavities. To correctly detect and classify them, a denoising is typically necessary for GPR images as the signal-to-noise ratio is low, and the various interfaces naturally present in the earth have a strong response. Both algorithms are based on a sparse convolutional coding model plus a low rank component. It is solved through an Alternating Direction Method of Multipliers (ADMM) framework. In order to take into account the presence of outliers and the artifacts caused by the acquisition, the second algorithm is based on the Huber norm instead of the classic L2 -norm. These algorithms are tested on a real dataset labeled by geophysicists. The results show the denoising efficiency of this approach, and in particular the robustness of the second algorithm.
@article{gallet:hal-04075772, title = {New Robust Sparse Convolutional Coding Inversion Algorithm for Ground Penetrating Radar Images}, author = {Gallet, Matthieu and Mian, Ammar and Ginolhac, Guillaume and Ollila, Esa and Stelzenmuller, Nickolas}, journal = {IEEE Transactions on Geoscience and Remote Sensing}, year = {2023}, doi = {10.1109/TGRS.2023.3268477}, keywords = {Ground Penetrating Radar ; Sparse Inversion ; Convolutive Model ; Robust methods ; Dictionaries ; Radar ; Signal to noise ratio ; Radar imaging ; Signal processing algorithms ; Shape ; Radar antennas}, url = {https://hal.science/hal-04075772}, }
Conference proceedings
2024
- IGARSSSupervised Classification for Analysis of Cryospheric Zones Using SAR Statistical TimeseriesLin-Kwong-Chon, Christophe, Gallet, Matthieu, Kaushik, Suvrat and 1 more authorIn IGARSS 2024 - 2024 IEEE International Geoscience and Remote Sensing Symposium 2024
This study explores machine learning for classifying X-band Synthetic Aperture Radar (SAR) monovariate time series from four cryospheric zones in the Mont-Blanc massif. We aim to classify ablation zones, accumulation zones, hanging glaciers, and ice aprons using log-cumulants and Dynamic Time Warping Barycentric Averaging. Our approach evaluates distances between time series and estimated reference centroids, employing HH and HV polarimetric channels. We propose an extension to this method by aggregating class membership probabilities from selected polarimetric combinations. Results are compared across polarimetric channels, revealing insights into classification performance.
- ICASSPRenyi Divergences Learning for explainable classification of SAR Image PairsGallet, Matthieu, Mian, Ammar, and Atto, AbdourrahmaneIn International Conference on Acoustics, Speech and Signal Processing (ICASSP) 2024
We consider the problem of classifying a pair of Synthetic Aperture Radar (SAR) images by proposing an explainable and frugal algorithm that integrates a set of divergences. The approach relies on a statistical framework that takes standard probability distributions into account for modelling SAR data. Then, by learning a combination of parameterized Renyi divergences and their parameters from the data, we are able to classify the pair of images with fewer parameters than regular machine learning approaches while also allowing an interpretation of the results related to the priors used. Experiments on real multi-class data demonstrate the virtues of the suggested method when compared to both Random Forest and Convolutional Neural Networks (CNN) classifiers, showing its resilience to disturbances such as polluted labels and variations in the percentage of training data.
@inproceedings{10448227, title = {Renyi Divergences Learning for explainable classification of SAR Image Pairs}, author = {Gallet, Matthieu and Mian, Ammar and Atto, Abdourrahmane}, booktitle = {International Conference on Acoustics, Speech and Signal Processing (ICASSP)}, year = {2024}, pages = {7445-7449}, keywords = {Training, Training data, Signal processing algorithms, Radar polarimetry,Probability distribution, Classification algorithms, Convolutional neural networks, SAR, Rényi divergence, explainable machine learning, classification}, doi = {10.1109/ICASSP48485.2024.10448227}, }
2023
- IGARSSCNN Classification of Wet Snow By Physical Snowpack Model LabelingGallet, Matthieu, Atto, Abdourrahmane, Trouvé, Emmanuel and 1 more authorIn International Geoscience and Remote Sensing Symposium (IGARSS 2023) Jul 2023
We propose a new approach for wet snow extent mapping in Synthetic Aperture Radar (SAR) images by using a convolutional neural network (CNN) designed to learn with respect to snowpack outputs from the state-of-the-art snow model Crocus. The CNN was trained to classify the wet snow conditions based on features extracted from the SAR images, using both the VV,VH channel and the ratio between these channels and those of a reference image in summer. One of the key points of this work is the comprehensive comparison we have made between the performance of the CNN method and other advanced statistical methods.We found that the CNN was able to achieve good accuracy in wet snow classification, and giving a complementary vision of the solutions obtained by other machine learning algorithms such as the Random Forest classifier. The results of this study demonstrate the potential of using CNNs and SAR images for wet snow classification and highlight the importance of using physical information model for training machine learning models in snow state identification, a domain where collecting ground truth is intricate due to the complexity of the snowpack moisture measurement systems.
@inproceedings{gallet:hal-04171432, title = {{CNN Classification of Wet Snow By Physical Snowpack Model Labeling}}, author = {Gallet, Matthieu and Atto, Abdourrahmane and Trouv{\'e}, Emmanuel and Karbou, Fatima}, booktitle = {{International Geoscience and Remote Sensing Symposium (IGARSS 2023)}}, address = {Pasadena, CA, United States}, year = {2023}, month = jul, keywords = {SAR ; Wet Snow ; Classification ; Convolutional Neural Networks}, hal_id = {hal-04171432}, hal_version = {v1}, url = {https://hal.science/hal-04171432}, } - GRETSIApprentissage explicable d’un ensemble de divergences pour la similarité inter-classe de données SARGallet, Matthieu, Atto, Abdourrahmane, Trouvé, Emmanuel and 1 more authorIn GRETSI, XXIXème Colloque Francophone de Traitement du Signal et des Images Aug 2023
In this work, we propose a method for classification of positive bivariate data using a set of divergences based on statistics, parametric marginal modeling and dependence modeling via copulas. These divergences are combined using neural network operators to determine the class of the input couple being considered. This approach is tested on a dataset of synthetic aperture radar (SAR) images from the PAZ satellite.
@inproceedings{gallet:hal-04184390, title = {{Apprentissage explicable d'un ensemble de divergences pour la similarit{\'e} inter-classe de donn{\'e}es SAR}}, author = {Gallet, Matthieu and Atto, Abdourrahmane and Trouv{\'e}, Emmanuel and Karbou, Fatima}, keywords = {SAR ; Similarity ; Divergence ; Copula ; Neural Network}, booktitle = {{GRETSI, XXIX{\`e}me Colloque Francophone de Traitement du Signal et des Images}}, address = {Grenoble, France}, organization = {{GRETSI}}, year = {2023}, month = aug, hal_id = {hal-04184390}, hal_version = {v1}, } - IGARSSTemporal Evolution of X and C Band SAR Backscattering in theMont-Blanc MassifKaushik, Suvrat, Gallet, Matthieu, Yan, Yajing and 3 more authorsIn International Geoscience and Remote Sensing Symposium (IGARSS 2023) Jul 2023
In this paper, two SAR image time series acquired by PAZ and Sentinel-1 satellites in 2020 (29 and 60 images respectively) are used to investigate surface changes of different ice/snow-covered areas in the Mont-Blanc massif. The evolution of the backscatter coefficient and several statistical parameters in both X and C band SAR images is analyzed on ice aprons, on valley glacier accumulation and ablation areas, and on ice-free areas. Dry and wet snow changes are observed and correlated with meteorological data (temperature at 4 different elevations and snow height) acquired by a weather station.
@inproceedings{kaushik:hal-04180662, title = {{Temporal Evolution of X and C Band SAR Backscattering in theMont-Blanc Massif}}, author = {Kaushik, Suvrat and Gallet, Matthieu and Yan, Yajing and Atto, Abdourrahmane and Ravanel, Ludovic and Trouv{\'e}, Emmanuel}, booktitle = {{International Geoscience and Remote Sensing Symposium (IGARSS 2023)}}, address = {Pasadena, CA, United States}, year = {2023}, month = jul, keywords = {SAR ; X/C bands ; Snow/Ice backscatter}, hal_id = {hal-04180662}, hal_version = {v1}, }
2022
- IGARSSClassification of GPR Signals via Covariance Pooling on CNN Features within a Riemannian FrameworkIn International Geoscience and Remote Sensing Symposium (IGARSS) Jul 2022
We consider the problem of classifying Ground Penetrating Radar (GPR) signals by using covariance matrices descriptors computed on convolutional features obtained from Mo-bileNetV2 Convolutional Neural Network (CNN) first layers. This approach allows to leverage the rich data representation obtained from CNNs and the low-dimensionality of secondorder statistics. Then the Riemannian geometry of covariance matrices is leveraged to improve classification rate. The proposed approach allows then to perform automatic classification of buried objects with few labeled data available. We also consider the scenario of an airbone radar and provide results at different elevations.
@inproceedings{gallet:hal-03726277, title = {Classification of GPR Signals via Covariance Pooling on CNN Features within a Riemannian Framework}, author = {Gallet, Matthieu and Mian, Ammar and Ginolhac, Guillaume and Stelzenmuller, Nickolas}, booktitle = {International Geoscience and Remote Sensing Symposium (IGARSS)}, address = {Kuala Lampur, Malaysia}, year = {2022}, month = jul, keywords = {GPR ; classification ; Convolutional Neural Networks ; Covariance matrix}, } - GRETSINouvel algorithme d’inversion robuste pour le RADAR GPRIn GRETSI, XXVIIIème Colloque Francophone de Traitement du Signal et des Images Sep 2022
Dans ce papier, nous proposons une nouvelle méthode d’inversion dans le but d’améliorer la détection d’objets enfouis. Pour cette détection, on utilise un RADAR GPR (Ground Penetrating Radar) qui émet une onde qui va traverser le sous sol et se réfléchir sur d’éventuels objets enterrés. A cause du mouvement du RADAR la réponse de ces objets a une forme d’hyperbole. L’approche proposée dans ce papier est basée sur un modèle convolutif avec dictionnaire et une matrice rang faible. Le problème d’optimisation utilise une norme d’Huber à la place de la norme classique pour avoir une meilleure robustesse au bruit qui est très présent dans une image GPR. Nous testons notre approche sur un jeu de données réelles fourni par la société Géolithe et montrons l’apport de la norme de Huber par rapport à la norme classique.
@inproceedings{gallet:hal-03726282, title = {Nouvel algorithme d'inversion robuste pour le RADAR GPR}, author = {Gallet, Matthieu and Mian, Ammar and Ginolhac, Guillaume and Stelzenmuller, Nickolas}, booktitle = {GRETSI, XXVIII{\`e}me Colloque Francophone de Traitement du Signal et des Images}, address = {Nancy, France}, organization = {{GRETSI}}, year = {2022}, month = sep, hal_id = {hal-03726282}, hal_version = {v1}, }
Datasets
2023
- ZenodoLSD4WSD : An Open Dataset for Wet Snow Detection with SAR Data and Physical LabellingGallet, Matthieu, Atto, Abdourrahmane, Karbou, Fatima and 1 more authorJul 2023
LSD4WSD: Learning SAR Dataset for Wet Snow Detection. The aim of this dataset is to provide a basis for automatic learning to detect wet snow. It is based on Sentinel-1 SAR satellite images acquired between August 2020 and August 2021 over the French Alps. It consists of 487157 samples of size 16 by 16 by 9 for training and 3668 for testing. For each sample, the associated label is obtained using the Crocus physical model.
@dataset{gallet_matthieu_2023_8111485, title = {LSD4WSD : An Open Dataset for Wet Snow Detection with SAR Data and Physical Labelling}, author = {Gallet, Matthieu and Atto, Abdourrahmane and Karbou, Fatima and Trouvé, Emmanuel}, year = {2023}, month = jul, publisher = {Zenodo}, version = {1.0.0}, doi = {10.5281/zenodo.10046730}, url = {https://doi.org/10.5281/zenodo.10046730}, }
Theses
2024
- Apprentissage automatique pour la caractérisation et l’analyse de la dynamique de la cryosphère par imagerie satellitaire radar.Gallet, MatthieuSep 2024
The analysis of the cryosphere is essential due to its socio-economic and environmental importance. It influences freshwater resources, supports the economic development of regions dependent on winter tourism, and plays a significant role in the global energy balance through the albedo effect. Globally, understanding the dynamics of the cryosphere is crucial for addressing current climate issues, as highlighted in the reports of the Intergovernmental Panel on Climate Change (IPCC).The explosion of satellite data in recent years offers opportunities for high spatial and temporal resolution monitoring of mountainous regions, particularly through data acquired by synthetic aperture radars (SAR). These radars allow for the observation of snowy and icy regions regardless of weather conditions and solar illumination. However, analyzing these data remains complex due to the presence of multiplicative noise (speckle) and irreversible geometric distortions, necessitating adapted analytical methods. The use of machine learning methods for SAR data analysis offers promising prospects for classifying snow-covered and glacier surfaces but requires approaches tailored to overcome the challenges related to data complexity and the quantity and quality of annotations.This thesis aims to address these challenges by focusing on several key aspects. Firstly, it identifies how to exploit C-band and X-band SAR data for snow and glacier classification by analyzing temporal behaviors as well as correlations between statistical image features and in situ or model-derived measurements.From this analysis, it proposes the creation of two annotated datasets for the study of the cryosphere and their use in classification tasks. These datasets consist of SAR images acquired by the Sentinel-1 and PAZ satellites, centered on the French Alps and the Mont-Blanc region, and annotated automatically with a snowpack simulation model (CROCUS) for wet snow and manually for glacier types. The creation of these datasets enables the analysis of wet snow detection by a set of supervised machine learning models and the understanding of discriminative features for this task. Given the variability in results and model performances, this thesis proposes an approach for aggregating classification models to improve the robustness of results in the face of inherent radar signal noise and annotation uncertainties, based on fuzzy logic.Finally, it proposes algorithmic solutions for effective and interpretable classification of SAR data, with a frugal approach in terms of parameters and computational time, combining the estimation of a set of probability distributions specific to SAR image features with a neural architecture for combining statistical distances or parameterizable divergences. Consequently, this work demonstrates the ability of machine learning methods to be effectively and interpretable adapted for SAR data analysis for the study of the cryosphere, considering the specifics of the data and annotations, and proposing algorithmic solutions for effective and interpretable classification.