AYANA

AYANA - 2025

2025Activity reportTeam‌AYANA

RNSR: null

Research‌‌ center Inria Centre at Université Côte d'Azur
Team‌ name: AI and Remote‌ Sensing on board for‌‌ the New Space

Creation of the Team: 2020‌ January 01

Each year,‌ Inria research teams publish‌‌ an Activity Report presenting their work and results‌ over the reporting period.‌ These reports follow a‌‌ common structure, with some optional sections depending on‌ the specific team. They‌ typically begin by outlining‌‌ the overall objectives and research programme, including the‌ main research themes, goals,‌ and methodological approaches. They‌‌ also describe the application domains targeted by the‌ team, highlighting the scientific‌ or societal contexts in‌‌ which their work is situated.

The reports then‌ present the highlights of‌ the year, covering major‌‌ scientific achievements, software developments, or teaching contributions. When‌ relevant, they include sections‌ on software, platforms, and‌‌ open data, detailing the‌ tools developed and how they are shared. A‌ substantial part is dedicated to new results, where‌ scientific contributions are described in detail, often with‌ subsections specifying participants and associated keywords.

Finally, the‌ Activity Report addresses funding, contracts, partnerships, and collaborations‌ at various levels, from industrial agreements to international‌ cooperations. It also covers dissemination and teaching activities,‌ such as participation in scientific events, outreach, and‌ supervision. The document concludes with a presentation of‌ scientific production, including major publications and those produced‌ during the year.

Keywords

Computer Science and Digital‌ Science

A2.3.1. Embedded systems
A5.9.2. Estimation, modeling
A5.9.4.‌ Signal processing over graphs
A5.9.6. Optimization tools
A9.2.1.‌ Supervised learning
A9.2.2. Unsupervised learning
A9.2.5. Bayesian methods‌
A9.2.6. Neural networks
A9.2.8. Deep learning
A9.3. Signal‌ processing
A9.12.1. Object recognition
A9.12.5. Object tracking and‌ motion analysis
A9.12.6. Object localization

1 Team members, visitors, external collaborators

Research‌ Scientist

Josiane Zerubia [Team leader, INRIA‌, Senior Researcher, HDR]

PhD Students‌

Louis Hauseux [UNIV COTE D'AZUR]
Priscilla‌ Indira Osa [UNIV GENES]

Administrative Assistant‌

Nathalie Nordmann [INRIA]

External Collaborators

Pierre‌ Charbonnier [Cerema Strasbourg, Head of the‌ ENDSUM team, HDR]
Zoltan Kato [‌UNIV SZEGED (HUN), full professor, Institut of‌ Informatics, HDR]
Gabriele Moser [UNIV‌ GENES, full professor, DITEN, HDR]‌
Martina Pastorino [UNIV GENES, assistant professor,‌ DITEN]
Sebastiano Serpico [UNIV GENES,‌ full professor, DITEN, HDR]

2 Overall‌ objectives

The AYANA AEx is an interdisciplinary project‌ using knowledge in stochastic modeling, image processing, artificial‌ intelligence, remote sensing and embedded electronics/computing. The aerospace‌ sector is expanding and changing ("New Space"). It‌ is currently undergoing a great many changes both‌ from the point of view of the sensors‌ at the spectral level (uncooled IRT, far ultraviolet,‌ etc.) and at the material level (the arrival‌ of nano-technologies or the new generation of "Systems‌ on Chips" (SoCs) for example), and from the‌ point of view of the carriers of these‌ sensors: high resolution geostationary satellites; Leo-type low-orbiting satellites;‌ or mini-satellites and industrial cube-sats in constellation. AYANA‌ will work on a large number of data,‌ consisting of very large images, having very varied‌ resolutions and spectral components, and forming time series‌ at frequencies of 1 to 60 Hz. For‌ the embedded electronics/computing part, AYANA will work in‌ close collaboration with specialists in the field located‌ in Europe, working at space agencies and/or for‌ industrial contractors.

3 Research program

3.1 FAULTS R‌ GEMS: Properties of faults, a key to realistic‌ generic earthquake modeling and hazard simulation

Decades of‌ research on earthquakes have yielded meager prospects for‌ earthquake predictability: we cannot predict the time, location and magnitude of a‌ forthcoming earthquake with sufficient‌ accuracy for immediate societal‌‌ value. Therefore, the best we can do is‌ to mitigate their impact‌ by anticipating the most‌‌ “destructive properties” of the largest earthquakes to come:‌ longest extent of rupture‌ zones, largest magnitudes, amplitudes‌‌ of displacements, accelerations of the ground. This topic‌ has motivated many studies‌ in the last decades.‌‌ Yet, despite these efforts, major discrepancies still remain‌ between available model outputs‌ and natural earthquake behaviors.‌‌ An important source of discrepancy is related to‌ the incomplete integration of‌ actual geometrical and mechanical‌‌ properties of earthquake causative faults in existing rupture‌ models. We first aim‌ to document the compliance‌‌ of rocks in natural permanent damage zones. These‌ data –key to earthquake‌ modeling– are presently lacking.‌‌ A second objective is to introduce the observed‌ macroscopic fault properties –compliant‌ permanent damage, segmentation, maturity–‌‌ into 3D dynamic earthquake models. A third objective‌ is to conduct a‌ pilot study aiming at‌‌ examining the gain of prior fault property and‌ rupture scenario knowledge for‌ Earthquake Early Warning (EEW).‌‌ This research project is partially funded by the‌ ANR Fault R Gems‌, whose PI is‌‌ Prof. I. Manighetti from Geoazur. Two successive‌ postdocs (Barham Jafrasteh‌ and Bilel Kanoun)‌‌ have worked on this research topic funded by‌ UCA-Jedi.

3.2 Probabilistic‌ models on graphs and‌‌ machine learning in remote sensing applied to natural‌ disaster response

We currently‌ develop novel probabilistic graphical‌‌ models combined with machine learning in order to‌ manage natural disasters such‌ as earthquakes, flooding and‌‌ fires. The first model introduces a semantic component‌ to the graph at‌ the current scale of‌‌ the hierarchical graph, and necessitates a new graph‌ probabilistic model. The quad-tree‌ proposed by Ihsen Hedhli‌‌ in AYIN team in 2016 is no longer‌ fit to resolve this‌ issue. Applications from urban‌‌ reconstruction or reforestation after natural disasters will be‌ achieved on images from‌ Pleiades optical satellites (provided‌‌ by the French Space Agency, CNES) and CosmoSkyMed‌ radar satellites (provided by‌ the Italian Space Agency,‌‌ ASI). This project is conducted in partnership with‌ the University of Genoa‌ (Prof. G. Moser and‌‌ Prof. S. Serpico) via the co-supervision of a‌ PhD student, financed by‌ the Italian government. The‌‌ PhD student, Martina Pastorino, has worked with‌ Josiane Zerubia and Gabriele‌ Moser in 2020 during‌‌ her double Master degree at both University Genoa‌ and IMT Atlantique. She‌ was a PhD Student‌‌ in co-supervision between University of Genoa DITEN (Prof.‌ Moser) and Inria‌ (Prof. Zerubia) from‌‌ November 2020 until December 2023, then a postdoc‌ in 2024, and is‌ now an assistant professor‌‌ at University of Genoa DITEN and external collaborator‌ of AYANA.

3.3 Marked‌ point process models for‌‌ object detection and tracking in temporal series of‌ high resolution images

The‌ model proposed by Paula‌‌ Craciun's PhD thesis in 2015 in AYIN‌ team, assumed the speed‌ of tracked objects of‌‌ interest in a series‌ of satellite images to be quasi-constant between two‌ frames. However, this hypothesis is very limiting, particularly‌ when objects of interest are subject to strong‌ and sudden acceleration or deceleration. The model we‌ proposed within AYIN team is then no longer‌ viable. Two solutions have been considered within AYANA‌ team : either a generalist model of marked‌ point processes (MPP), assuming an unknown and variable‌ velocity, or a multimodel of MPPs, simpler with‌ regard to the velocity that can have a‌ very limited amount of values (i.e., quasi-constant velocity‌ for each MPP). The whole model has been‌ redesigned, and is to be tested with data‌ from a constellation of small satellites (CO3D launched‌ by Vega-C in July 2025), where the objects‌ of interest can be for instance "speed-boats" or‌ "go-fast cars". Some comparisons with deep learning based‌ methods belonging to Airbus Defense and Space (Airbus‌ DS) are planned at Airbus DS. Then this‌ new model should be brought to diverse platforms‌ (ground based or on-board). The modeling and ground-based‌ application part related to object detection has been‌ studied within the PhD thesis of Jules Mabon‌. Furthermore, the object-tracking methods have been proposed‌ by AYANA as part of a postdoctoral project‌ (Camilo Aguilar-Herrera). Finally, the on-board version‌ will be developed by Airbus DS, and the‌ company Erems for the onboard hardware. This project‌ is financed by Bpifrance within the LiChIE contract.‌ This research project has started in 2020 for‌ a duration of 6 years.

4 Application domains‌

Our research is applied within all Earth observation‌ domains such as: urban planning, precision farming, natural‌ disaster management, geological features detection, geospatial mapping, and‌ security management.

5 Highlights of the year

5.1‌ Awards

For more information, see the AYANA news‌ webpage.

Louis Hauseux was finalist of the‌ Prix Pierre Laffitte 2025.
Martina Pastorino received‌ the following prizes:
- IEEE GRSS European Best PhD‌ Thesis Prize of the IEEE Geoscience and Remote‌ Sensing Society (GRSS), 2025.
- IEEE GRSS PhD Thesis‌ Prize issued by the Italian Chapters of the‌ IEEE Geoscience and Remote Sensing Society (GRSS), 2025.‌
- Premio Internazionale Galileo Galilei, Rotary District 2032‌ (Italy), 2025
Josiane Zerubia received the following awards:‌
- University of Bristol granted her the title of‌ “Bristol Benjamin Meaker Distinguished Visiting Professor”.‌
- She was appointed as an IEEE GRSS Distinguished‌ Lecturer for 2025 – 2027.
- She was promoted‌ to IEEE Life Fellow starting January 2026.

6‌ Latest software developments, platforms, open data

6.1 Latest‌ software developments

6.1.1 MPP & CNN for object‌ detection in remotely sensed images

Name:
Marked Point‌ Processes and Convolutional Neural Networks for object detection‌ in remotely sensed images
Keywords:
Detection, Satellite imagery‌
Functional Description:
Implementation of the work presented in:‌ "CNN-based energy learning for MPP object detection in‌ satellite images" Jules Mabon, Mathias Ortner, Josiane Zerubia‌ In Proc. 2022 IEEE International Workshop on Machine‌ Learning for Signal Processing (MLSP) and "Point process and CNN for small‌ objects detection in satellite‌ images" Jules Mabon, Mathias‌‌ Ortner, Josiane Zerubia In Proc. 2022 SPIE Image‌ and Signal Processing for‌ Remote Sensing XXVIII
URL:‌‌
https://github.com/Ayana-Inria/MPP_CNN_RS_object_detection
Contact:
Josiane Zerubia
Partner:
Airbus Defense and‌ Space

6.1.2 FCN and‌ Fully Connected NN for‌‌ Remote Sensing Image Classification

Name:
Fully Convolutional Network‌ and Fully Connected Neural‌ Network for Remote Sensing‌‌ Image Classification
Keywords:
Satellite imagery, Classification, Image segmentation‌
Functional Description:

Code related‌ to the paper:

M.‌‌ Pastorino, G. Moser, S. B. Serpico, and J.‌ Zerubia, "Fully convolutional and‌ feedforward networks for the‌‌ semantic segmentation of remotely sensed images," 2022 IEEE‌ International Conference on Image‌ Processing, 2022,
URL:
https://github.com/Ayana-Inria/FCN-FFNET_RS-semantic-segmentation‌‌
Contact:
Josiane Zerubia
Partner:
University of Genoa, DITEN,‌ Italy

6.1.3 Stabilizer for‌ Satellite Videos

Keywords:
Satellite‌‌ imagery, Video sequences
Functional Description:
A python-implemented stabilizer‌ for satellite videos. This‌ code was used to‌‌ produce the object tracking results shown in: C.‌ Aguilar, M. Ortner and‌ J. Zerubia, "Adaptive Birth‌‌ for the GLMB Filter for object tracking in‌ satellite videos," 2022 IEEE‌ 32st International Workshop on‌‌ Machine Learning for Signal Processing (MLSP), 2022, pp.‌ 1-6
URL:
https://github.com/Ayana-Inria/satellite-video-stabilization
Contact:‌
Josiane Zerubia
Partner:
Airbus‌‌ Defense and Space

6.1.4 GLMB filter with History-based‌ Birth

Name:
Python GLMB‌ filter with History-based Birth‌‌
Keywords:
Multi-Object Tracking, Object detection
Functional Description:
Implementation‌ of the work presented‌ in: C. Aguilar, M.‌‌ Ortner and J. Zerubia, "Adaptive Birth for the‌ GLMB Filter for object‌ tracking in satellite videos,"‌‌ 2022 IEEE 32st International Workshop on Machine Learning‌ for Signal Processing (MLSP),‌ 2022, pp. 1-6
URL:‌‌
https://github.com/Ayana-Inria/GLMB-adaptive-birth-satellite-videos
Contact:
Josiane Zerubia
Partner:
Airbus Defense and‌ Space

6.1.5 Automatic fault‌ mapping using CNN

Keyword:‌‌
Detection
Functional Description:
Implementation of the work published‌ in: Bilel Kanoun, Mohamed‌ Abderrazak Cherif, Isabelle Manighetti,‌‌ Yuliya Tarabalka, Josiane Zerubia. An enhanced deep learning‌ approach for tectonic fault‌ and fracture extraction in‌‌ very high resolution optical images. ICASSP 2022 -‌ IEEE International Conference on‌ Acoustics, Speech, & Signal‌‌ Processing, IEEE, May 2022, Singapore/Hybrid, Singapore.
URL:
https://github.com/Ayana-Inria/Automatic_fault_mapping_using_CNN‌
Contact:
Josiane Zerubia
Partner:‌
Géoazur

6.1.6 RFS-filters for‌‌ Satellite Videos

Keywords:
Detection, Target tracking
Functional Description:‌
Implementation of the works‌ published in Camilo Aguilar,‌‌ Mathias Ortner, Josiane Zerubia. Enhanced GM-PHD filter for‌ real time satellite multi-target‌ tracking. ICASSP 2023 –‌‌ IEEE International Conference on Acoustics, Speech, and Signal‌ Processing, Jun 2023, Rhodes,‌ Greece.
URL:
https://github.com/Ayana-Inria/RFS-filters-for-satellite-videos
Contact:‌‌
Josiane Zerubia
Partner:
Airbus Defense and Space

6.1.7‌ AYANet

Name:
AYANet: A‌ Gabor Wavelet-based and CNN-based‌‌ Double Encoder for Building Change Detection in Remote‌ Sensing
Keywords:
Gabor wavelet,‌ Convolutional Neural Network, Building‌‌ Change Detection, Remote Sensing
Functional Description:
The official‌ implementation of AYANet: A‌ Gabor Wavelet-based and CNN-based‌‌ Double Encoder for Building Change Detection in Remote‌ Sensing (ICPR 2024) Link‌ to the paper :‌‌ https://hal.science/hal-04675243
URL:
https://team.inria.fr/ayana/
Contact:
Josiane Zerubia

6.1.8 GabFormer‌

Name:
Gabor Feature Network‌ for Transformer-based Building Change‌‌ Detection Model in Remote Sensing
Keywords:
Transformer, Gabor‌ wavelet, Building Change Detection,‌ Remote Sensing, Image analysis‌‌
Functional Description:
The official‌ implementation of Gabor Feature Network for Transformer-based Building‌ Change Detection Model in Remote Sensing (ICIP 2024)‌ Link to the paper: https://hal.science/hal-04619245
URL:
https://team.inria.fr/ayana/
Contact:‌
Josiane Zerubia

6.1.9 PP-EBM

Name:
Combining Convolutional Neural‌ Networks and Point Process for object detection
Keywords:‌
Convolutional Neural Network, Point Process, Energy based model,‌ Remote Sensing
Functional Description:

This code was used‌ to produce the results shown in:

Mabon, J.,‌ Ortner, M., Zerubia, J. Learning Point Processes and‌ Convolutional Neural Networks for Object Detection in Satellite‌ Images. Remote Sens. 2024, 16, 1019. https://doi.org/10.3390/rs16061019
URL:‌
https://team.inria.fr/ayana/
Contact:
Josiane Zerubia
Partner:
Airbus Defense and‌ Space

6.1.10 HGP-K2

Name:
HypergraphPercol-K2
Keywords:
Clustering, Point‌ cloud, Hypergraph, Percolation
Functional Description:
The software consists‌ essentially in one clustering function called HypergraphPercol. See‌ readme in the GitHub. It generalizes and gives‌ better results than HDBSCAN for point cloud clustering.‌
URL:
https://github.com/Ayana-Inria/HypergraphPercol_K2
Contact:
Josiane Zerubia

6.1.11 CFC-CRF

Name:‌
Cluster level Fully Connected CRF
Keywords:
Hyperspectral imagery,‌ Multiresolution fusion, Semantic Segmentation, FCN, CRF
Functional Description:‌
The official implementation of the CRF in CFC-MCRF:‌ Multiresolution fusion and segmentation of hyperspectral and panchromatic‌ remote sensing images with deep learning and CRFs‌ presented at EUSIPCO'25.
URL:
https://github.com/martinapastorino/Semisupervised-CRF
Contact:
Martina Pastorino‌

6.1.12 BAS-UAV

Name:
Burnt area segmentation from satellite‌ and UAV images
Keywords:
Probabilistic fusion, Probabilistic graphical‌ models, Deep learning, Multiresolution imagery, Semantic Segmentation, Forest‌ fires
Functional Description:
The official implementation of Probabilistic‌ Fusion Framework Based on Fully Convolutional Networks and‌ Graphical Models for Burnt Area Detection from Multiresolution‌ Satellite and UAV Imagery accepted for publication in‌ IEEE TGRS
URL:
https://github.com/Ayana-Inria/BAS_UAV_satellite_fusion
Contact:
Josiane Zerubia

7‌ New results

7.1 Probabilistic graphical models and convolutional‌ networks for the semantic segmentation

Participants: Martina Pastorino‌, Josiane Zerubia.

External collaborators: Gabriele Moser‌ [University of Genoa, DITEN dept., Professor], Sebastiano‌ Serpico [University of Genoa, DITEN dept., Professor].‌

Keywords: image processing, stochastic models, deep learning, fully‌ convolutional networks.

Semantic segmentation, also known as spatially‌ dense image classification, plays a crucial role in‌ image analysis, bridging the fields of image processing‌ and machine learning. It has wide applications, ranging‌ from land cover mapping in Earth observation to‌ medical diagnostics using biomedical images, fault detection in‌ industrial imagery, etc. This work focuses on the‌ mathematical connections between two pivotal families of methodological‌ approaches: probabilistic graphical models (PGM) and deep learning‌ (see Fig. 1), and explores the potential‌ of their integration for semantic segmentation tasks. After‌ providing a comprehensive overview of state-of-the-art techniques from‌ both families, the work highlights recent developments that‌ combine these approaches, either through theoretical equivalence or‌ direct integration. Examples of results are provided for‌ renowned benchmark datasets in computer vision and remote‌ sensing and have been accepted for publication in‌ the IEEE Signal Processing Magazine 4.

Figure 1‌: Structural analogy between an encoder-decoder based neural‌ network (left) and a quadtree topological structure (right).

7.2 A Probabilistic Fusion‌ Framework for Burnt zones‌ mapping from multimodal satellite‌‌ and UAV imagery

Participants: Martina Pastorino, Josiane‌ Zerubia.

External collaborators:‌ Gabriele Moser [University of‌‌ Genoa, DITEN dept., Professor], Sebastiano Serpico [University‌ of Genoa, DITEN dept.,‌ Professor], Fabien Guerra‌‌ [Inrae, ERCOVER dept., University of Aix-Marseille].

Keywords:‌ image processing, stochastic models,‌ deep learning, satellite images,‌‌ UAV imagery.

Semantic segmentation has also been addressed‌ in the framework of‌ natural disaster management applications,‌‌ specifically to detect zones affected by forest fires.‌ In this context, on‌ one hand, satellite images‌‌ provide a synoptic view, although with limited spatial‌ detail and possibly being‌ affected by haze and‌‌ smoke. On the other hand, images collected using‌ unmanned aerial vehicles (UAVs,‌ drones) achieve spatial resolutions‌‌ of a few centimeters and are mostly unaffected‌ by smoke and haze,‌ but intrinsically focus on‌‌ a smaller geographical coverage. The proposed approach addresses‌ the fusion of these‌ two data modalities in‌‌ the application to fire scar mapping. The resulting‌ multiresolution fusion task is‌ especially challenging (see Fig.‌‌ 2) because the ratio between the involved‌ spatial resolutions is extremely‌ large (e.g.,‌‌ 10 m from Sentinel-2 vs 5 cm from‌ an UAV) – a‌ situation that is normally‌‌ not addressed by multiresolution schemes from the state‌ of the art. Two‌ novel Bayesian approaches, combining‌‌ deep fully convolutional networks, ensemble learning, decision fusion,‌ and hierarchical PGMs have‌ been proposed. The experimental‌‌ validation, conducted in collaboration with forest science experts‌ from Inrae (Institut national‌ de recherche pour l’agriculture,‌‌ l’alimentation et l’environnement) and with regard to two‌ case studies associated with‌ wildfires in Provence, have‌‌ suggested the capability of the developed approaches to‌ benefit from both remote‌ sensing image modalities to‌‌ achieve high detection performance. These proposed methods have‌ been accepted for publication‌ in TGRS 21.‌‌

Figure 2.a — Figure‌‌ 2: Input images, ground truth, and classification‌ results: (a) drone image‌ at 2 cm of‌‌ spatial resolution, (b) NDVI Sentinel-2 image at 10‌ m of spatial resolution,‌ (c) the ground truth‌‌ with the same resolution of the drone image,‌ (d) testing ground truth,‌ and (e) the classification‌‌ result of the proposed methods. Class legend: burnt‌ (red) and non-burnt (white).‌

Figure 2.b — Figure‌‌ 2: Input images, ground truth, and classification‌ results: (a) drone image‌ at 2 cm of‌‌ spatial resolution, (b) NDVI Sentinel-2 image at 10‌ m of spatial resolution,‌ (c) the ground truth‌‌ with the same resolution of the drone image,‌ (d) testing ground truth,‌ and (e) the classification‌‌ result of the proposed methods. Class legend: burnt‌ (red) and non-burnt (white).‌

7.3 Multiresolution Fusion and‌‌ Segmentation of Hyperspectral and Panchromatic Remote Sensing Images‌ with deep learning and‌ probabilistic graphical models

Participants:‌‌ Martina Pastorino, Josiane Zerubia.

External collaborators:‌ Gabriele Moser [University of‌ Genoa, DITEN dept., Professor]‌‌, Sebastiano Serpico [University of Genoa, DITEN dept.,‌ Professor].

Keywords: image‌ processing, stochastic models, deep‌‌ learning, satellite images, hyperspectral‌ imagery.

This work proposes a supervised method for‌ the joint classification and fusion of multiresolution panchromatic‌ and hyperspectral data based on the combination of‌ probabilistic graphical models (PGMs) and deep learning methods.‌

The idea is to exploit the spatial and‌ spectral information contained in panchromatic and hyperspectral images‌ at different resolutions with the aim to generate‌ a classification map at the spatial resolution of‌ the panchromatic channel, while exploiting the richness of‌ the spectral information provided by the hyperspectral channels.‌

The proposed technique is based on deep learning,‌ with FCN-type (Fully Convolutional Networks) or vision-transformer (ViTs)‌ architectures, and PGMs, through the definition of a‌ hierarchical conditional random field (CRF) approximating the behavior‌ of the ideal fully connected CRF in a‌ computationally tractable manner. The neural architecture (either based‌ on FCNs or ViTs) aims to integrate hyperspectral‌ and panchromatic data at the corresponding spatial resolution‌ and generate posterior probability estimates, while the CRF‌ incorporates information associated with not only local but‌ also long-distance spatio-spectral relationships.

The algorithm has been‌ experimentally validated with PRISMA data in the framework‌ of a project with the Italian Space Agency‌ with promising results (see Fig. 3). The‌ proposed method has been presented at EUSIPCO'25 9‌ and have been submitted (in revision) to TGRS‌ 22.

Figure 3.a — Figure 3: Test‌ ground truths and classification maps for the test‌ tiles in the PRISMA dataset (Pavia, Italy). Product‌ processed under a license of the Italian Space‌ Agency (ASI); Original PRISMA Product - ©ASI -‌ (2022).

Figure 3.b — Figure 3: Test‌ ground truths and classification maps for the test‌ tiles in the PRISMA dataset (Pavia, Italy). Product‌ processed under a license of the Italian Space‌ Agency (ASI); Original PRISMA Product - ©ASI -‌ (2022).

7.4 Joint Deep-learning-based Gap-filling and Clustering of‌ Satellite Image Time Series

Participants: Priscilla Indira Osa‌, Josiane Zerubia.

External collaborators: Gabriele Moser‌ [University of Genoa, DITEN dept., Professor], Sebastiano‌ Serpico [University of Genoa, DITEN dept., Professor].‌

Keywords: satellite image time series, gap filling, clustering,‌ unsupervised classification, deep learning, joint optimization.

Unsupervised pixel-wise‌ classification in optical satellite image time series (SITS‌ clustering) typically deals with the problem of missing‌ values due to the cloud cover or shadows‌ during the data acquisition. This problem is conventionally‌ addressed by applying a gap-filling method before applying‌ the clustering algorithm to the data. This inevitably‌ makes the final results influenced by the choice‌ of the two above mentioned methods. In order‌ to minimize the effect coming from the combination‌ of arbitrary models addressing the two elements separately,‌ a model is developed to solve gap-filling and‌ clustering problems jointly, taking inspiration from 19,‌ 24. Figure 4 shows the overall architecture‌ of the proposed method. The gap-filling of the‌ satellite image time series is done by the‌ Generator, while the representation learning and cluster assignment‌ are performed by the Autoencoder. During the training, the adversarial strategy is‌ implemented between the Discriminator‌ and the remaining parts‌‌ of the model.

Figure 4‌: The overall architecture‌‌ of the proposed model.

The experimental validation was‌ done by classifying a‌ SITS of vegetation index‌‌ called Fraction of Photosynthetically Active Radiation (FPAR) with‌ 1 km of spatial‌ resolution in 2021 in‌‌ the region of Italy, into winter and summer‌ crops. The results were‌ compared with several gap-filling‌‌ algorithms (mean-value imputation, zero imputation, linear interpolation, and‌ cubic spline interpolation), and‌ different deep clustering techniques‌‌ (Autoencoder + k-means, Autoencoder + Gaussian Mixture Model‌ (GMM), DEC 24,‌ DTJC 17, CRLI‌‌ 19). A sample of qualitative comparison, shown‌ in Fig. 5,‌ illustrates the efficiency of‌‌ the proposed model.

Figure‌‌ 5: The qualitative results of the comparison‌ between the proposed and‌ existing methods.

The study‌‌ was conducted within the RETURN Extended Partnership and‌ received funding from the‌ European Union Next-GenerationEU. The‌‌ work was presented at IEEE International Geoscience and‌ Remote Sensing Symposium8‌ (IGARSS) 2025 in Brisbane‌‌ (Australia) and IEEE European Signal Processing Conference (EUSIPCO)‌ 2025 in Palermo (Italy)‌ 7.

7.5 Generalization‌‌ of Single-Linkage: Hypergraphs and Percolation Analysis

Participants: Louis‌ Hauseux, Konstantin Avrachenkov‌, Josiane Zerubia.‌‌

Keywords: hierarchical clustering, density-based clustering, geometric graphs, high-order‌ interactions, simplicial complexes, percolation.‌

Single-Linkage 14, 16‌‌ is a classical clustering algorithm, widely used for‌ its simplicity and theoretical‌ properties 11, 23‌‌. However, it suffers from the "chaining effect"‌ (which consists of the‌ tendency for clusters to‌‌ grow into long shapes, `chains'). Its modern robust‌ variant is the core‌ of HDBSCAN 12,‌‌ 20, the current state-of-the-art in density-based clustering.‌

We show how to‌ naturally generalize this approach‌‌ by introducing higher-order interactions. Instead of simple‌ graphs, we utilize hypergraphs‌ (specifically Čech complexes) and‌‌ define a stricter notion of connectivity called $K‌$ -polyhedra. We proved that‌ these $K$ -polyhedra correspond‌‌ to the high-density clusters of the $K$ -Nearest‌ Neighbors density estimator. Practically,‌ our algorithm identifies a‌‌ "Minimum $K$ -Tree", generalizing the use of the‌ Minimum Spanning Tree in‌ the Single-Linkage algorithm.

In‌‌ 18, we focused on theoretically quantifying the‌ performance of this family‌ of algorithms. We identified‌‌ percolation13 as the key phenomenon governing clustering‌ quality. We introduced a‌ novel performance index: the‌‌ percolation rate. As illustrated in Fig. 6‌, the percolation transition‌ separates a "detection phase"‌‌ from a "recovery phase". A faster percolation rate‌ implies a sharper transition‌ and a better ability‌‌ to distinguish neighboring high-density clusters.

Figure‌ 6: Percolation probability $p_{\infty} (r‌)$ showing the phase transition. The percolation rate‌ measures the gap between the detection of structures‌ and the complete recovery of data. A steeper‌ slope indicates better performance. Here the curve is‌ the percolation probability of the geometric graphs (‌ $K = 1$ ) in $ℝ^{2}$ .‌

We demonstrated that our $K$ -polyhedra approach (‌ $K \geq 2$ ) achieves a significantly higher‌ percolation rate compared to the state-of-the-art $K$ -Robust‌ Single-Linkage used in HDBSCAN (see Tab. 1).‌

Table 1: Percolation rate in $ℝ^{2}$ .‌ Our proposed $K$ -polyhedra (right handside) shows a‌ faster transition than the standard Robust Single-Linkage (left‌ handside), validating its superior separation capabilities.
Clustering	$K‌$ -Robust Single-Linkage	$K$ -polyhedra (Ours)
$K = 1‌$	$1.12 / 1.75 =‌$ 0.64	$1 . 04 / 1 . 67‌ \approx$ 0.62
$K = 2$	$1 . 12‌ / 1 . 85 \approx$ 0.61	$2 .‌ 23 / 3 . 05 \approx$ 0.73
$K‌ = 3$	$1 . 27 / 2 .‌ 11 \approx$ 0.60	$3 . 40 / 4‌ . 48 \approx$ 0.76

These results were presented‌ at the Complex Networks 2024 conference 18 and‌ the ACM Sigmetrics Student Research Competition6 (2‌nd prize). A comprehensive journal article about these‌ results has been accepted in Applied Network Science‌3.

7.6 Generalized Frangi Filter: Multimodal Fusion‌ for Crack Extraction

Participants: Louis Hauseux, Josiane‌ Zerubia.

External collaborators: Pierre Charbonnier [Cerema Strasbourg,‌ Research scientist], Philippe Foucher [Cerema Strasbourg, Research‌ scientist], Raphaël Antoine [Cerema Normandie-Centre, Research scientist]‌.

Keywords: generalized Frangi filter, geometric graphs, crack‌ extraction, multimodal fusion, LiDAR.

Detecting linear anomalies, such‌ as cracks on civil engineering structures or geological‌ faults, is a critical task. The classical Frangi‌ filter 15 enhances tubular structures using the Hessian‌ matrix but operates pixel-wise. It is insufficient to‌ exhibit the whole linear anomaly network.

We developed‌ a Generalized Frangi Filter that shifts the analysis‌ from pixels to edges (pairs of pixels). Instead‌ of classifying isolated points, we evaluate the likelihood‌ that two pixels belong to the same tubular‌ structure. This allows us to incorporate higher-order geometric‌ attributes such as the alignment angle.

Since 5‌, a major contribution at the end of‌ 2025 has been made: the extension of this‌ framework to multimodal data fusion. By combining optical‌ information (Intensity) with topographic data (‌LiDAR Range/Depth), we significantly improve robustness. As‌ illustrated in Fig. 7, fusing these modalities‌ allows us to recover crack skeletons even when‌ one modality is heavily corrupted (e.g.,‌ by speckle or Gaussian noise), whereas single-modality approaches‌ fail.

Figure 7: Robustness‌‌ through Multimodal Fusion on the FIND dataset. From‌ left to right, top‌ to bottom: Noisy Intensity‌‌ input, noisy Range input, then five steps of‌ our Generalized Frangi framework,‌ and last one: the‌‌ response successfully recovering the crack network.

From these‌ pairwise responses, we construct‌ a weighted graph and‌‌ extract the crack network using our generalized Single-Linkage‌ algorithm (see Section 7.5‌). We validated this‌‌ approach on the "Vaches Noires" landslide 5 and‌ the FIND dataset 25‌. Quantitative comparisons against‌‌ state-of-the-art deep learning methods (e.g., CrackSegDiff‌) demonstrate that our‌ unsupervised topological approach provides‌‌ superior robustness with respect to noise and domain‌ shifts, without requiring training‌ data.

The first part‌‌ of this work was presented at the GRETSI‌ 2025 conference 5.‌ The Multi-Modal robust version‌‌ (with comparison to the state-of-the-art Deep Learning method)‌ will be submitted to‌ EUSIPCO 2026.

7.7 3D‌‌ LiDAR Instance Segmentation with Geometric Priors: The HGP-Clusterer‌

Participants: Louis Hauseux,‌ Josiane Zerubia.

Keywords:‌‌ point cloud segmentation, anomaly detection, geometric priors, 3D‌ model, 3D LiDAR.

While‌ the density-based framework (see‌‌ Section 7.5) provides a robust theoretical basis‌ for clustering, real-world 3D‌ LiDAR scenes present unique‌‌ challenges such as variable density and occlusions. To‌ address these problems, we‌ developed the HGP-clusterer (Hypergraph-Percolation‌‌ Clusterer), a high-performance library able to build the‌ $K$ -polyhedra hierarchy on‌ large-scale point clouds.

An‌‌ important improvement of our research work in 2025‌ is the development of‌ a Guided Hierarchical Clustering‌‌ method. Standard algorithms like HDBSCAN rely on a‌ generic "stability" criterion to‌ cut the dendrogram. However,‌‌ we demonstrated that we can inject geometric priors‌ directly during the hierarchical‌ tree exploration. Instead of‌‌ maximizing the lifetime of a cluster, we implemented‌ a customized loss function.‌ This function evaluates whether‌‌ a cluster should be split or merged based‌ on its consistency with‌ expected geometric properties (‌‌e.g., bounding box alignment, volume) rather than‌ just density persistence.

We‌ applied this framework to‌‌ a change detection synthetic benchmark provided by Marie‌ Aspro (working with Naval‌ Group). The goal is‌‌ to automatically identify "anomalies" (objects present in the‌ LiDAR scan but missing‌ from the reference 3D‌‌ model).

We utilized the 3D model as a‌ geometric prior to guide‌ the HGP-clusterer. During the‌‌ tree descent, the algorithm filters out points matching‌ the known environment structure‌ and isolates unexpected dense‌‌ components. To handle fragmentation caused by occlusions (‌e.g., a single‌ object split into multiple‌‌ segments), we added a topological post-processing step that‌ fuses adjacent anomaly clusters‌ based on bounding-box intersections.‌‌

Figure 8‌: Anomaly detection on‌‌ a shipyard synthetic benchmark.‌ The algorithm uses the 3D model as a‌ geometric prior to filter out valid objects (red)‌ and successfully isolates foreign objects (other colors than‌ red and gray), despite low point density background‌ (gray). ©Naval Group.

Figure 8 illustrates the results:‌ the algorithm successfully detects and segments the target‌ anomalies while ignoring "out-of-field" noise. This demonstrates that‌ injecting weak geometric supervision into the clustering process‌ offers a robust, training-free alternative to deep learning‌ for industrial applications.

We would like to thank‌ Marie Aspro (Inria Startup Studio) for providing us‌ with this benchmark.

8 Bilateral contracts and grants‌ with industry

8.1 Bilateral contracts with industry

8.1.1‌ LiChIE contract with Airbus Defense and Space funded‌ by Bpifrance

Participants: Josiane Zerubia.

External collaborators:‌ Mathias Ortner [Airbus Defense and Space, Senior Data‌ Scientist].

Automatic object detection and tracking on‌ sequences of images taken from various constellations of‌ satellites. This contract covered one PhD (Jules Mabon‌ from October 2020 until January 2024), one postdoc‌ (Camilo Aguilar Herrera from January 2021 until September‌ 2022), and one research engineer (Louis Hauseux from‌ February until April 2023). Josiane Zerubia worked on‌ this contract during six years (2020-2025).

9 Partnerships‌ and cooperations

9.1 International research visitors

9.1.1 Visits‌ to international teams

Research stays abroad

Josiane Zerubia‌

Visited institution:
University of Bristol
Country:
United Kingdom‌
Dates:
July 1-31, 2025
Context of the visit:‌
Josiane Zerubia visited in July 2025 Prof. Alin‌ Achim team in the School of Computer Science‌ at the University of Bristol.
Mobility program/type of‌ mobility:
Supported by Bristol Benjamin Meaker Distinguished Visiting‌ Professor award.

9.2 National initiatives

ANR FAULTS‌ R GEMS

Participants: Josiane Zerubia.

External collaborators:‌ Isabelle Manighetti [OCA, Géoazur, Senior Physicist].

AYANA‌ team is part of the ANR project FAULTS‌ R GEMS (2017-2025, PI Geoazur) dedicated to realistic‌ generic earthquake modeling and hazard simulation.

9.3 Prospecting‌ industrial partners

Participants: Josiane Zerubia.

Josiane Zerubia‌ visited the following potential industrial partners, to set‌ up national contracts with Inria (e.g.,‌ Defi Inria/..., AID proposals):
- Visits to Acri-ST: Toulouse‌ (December 2025), Grasse (May 2025).
- Visits to: Airbus‌ DS Toulouse, February and December 2025.

10 Dissemination‌

Participants: Louis Hauseux, Priscilla Indira Osa,‌ Martina Pastorino, Josiane Zerubia.

10.1 Promoting‌ scientific activities

10.1.1 Scientific events: selection

Member of‌ the conference program committees

Josiane Zerubia was a‌ member of the conference program committee of SPIE‌ Artificial Intelligence and Image and Signal Processing for‌ Remote Sensing'25 in Madrid (Spain), September 2025.‌

Reviewer

Martina Pastorino reviewed the following international conferences:‌ IEEE IGARSS'25 (Student Paper Competition track), and‌ ECML-PKDD'25.
Josiane Zerubia reviewed the following international‌ conferences: IEEE-EURASIP EUSIPCO'25, SPIE Artificial Intelligence and‌ Image and Signal Processing for Remote Sensing'25,‌ IEEE EMBC'25, IEEE ICASSP'25, IEEE IGARSS'25‌, IEEE ICIP'25.

10.1.2 Journal

Member of‌ the editorial boards

Josiane Zerubia has been a‌ member of the editorial board of Foundation and Trends in Signal Processing‌ (2007- ).

Reviewer -‌ reviewing activities

Martina Pastorino‌‌ reviewed the following international journals: Nature Communications,‌ Neurocomputing, IEEE SPM‌, IEEE TGRS,‌‌ IEEE TIP, IEEE TCSVT, IEEE GRSL‌, IEEE JSTARS,‌ and MDPI Remote Sensing‌‌.

10.1.3 Invited talks

Louis Hauseux gave the‌ following seminars:
- Seminar in‌ Valeo AI team in‌‌ Paris, November 2025.
- MATHNET/DYOGENE seminar at Inria-Paris‌ center, March 2025.
Martina‌ Pastorino was invited to‌‌ present her work at the hyperspectral SFPT-colloquium in‌ ACRI-ST in Grasse, May‌ 2025.
Josiane Zerubia gave‌‌ the following invited talks:
- A plenary talk at‌ SYNASC'25 in Timișoara (Romania)‌ and chaired an AI‌‌ session at the same conference, September 2025.
- Two‌ plenaries at Bristol University‌, one at the‌‌ School of Computer Science, the other one‌ at Wills Memorial Building‌, July 2025.

10.1.4‌‌ Leadership within the scientific community

Josiane Zerubia is‌ a Fellow of IEEE‌ (2002- ), EURASIP (2019-‌‌ ) and IAPR (2020-).
Josiane Zerubia is a‌ member of the Teaching‌ Board of the Doctoral‌‌ School STIET at University of Genoa, Italy (2018-‌ ).
Josiane Zerubia is‌ a senior member of‌‌ IEEE WISP Committee (2024-2025).
Josiane Zerubia is part‌ of the EURASIP Fellows‌ Selection Committee (2025-2027).
Josiane‌‌ Zerubia did an evaluation for new member selection‌ at the Science Academy,‌ Istanbul (Turkey, Bilim Akademisi),‌‌ July 2025.

10.1.5 Scientific expertise

Josiane Zerubia did‌ consulting for CAC (Cosmeto‌ Azur Consulting), December 2025.‌‌

10.1.6 Research administration

Louis Hauseux is in charge‌ of the C@fé ADSTIC‌ since 2023.
Josiane Zerubia‌‌ is a member of Comité des Equipes Projets‌ (CEP) at Inria-Centre d'Université‌ Côte d'Azur.
Josiane Zerubia‌‌ is in charge of the aerospace research field‌ for the Defense and‌ Security mission at Inria,‌‌ headed by Frédérique Segond since January 2023. She‌ was invited to
10.2‌ Teaching - Supervision -‌‌ Juries - Educational and pedagogical outreach
10.2.1 Supervision‌
- Louis Hauseux gave the‌ following courses to Master‌‌ students:
  - Teaching course and practicals with Konstantin Avrachenkov‌ and Rémy Sun:‌ master course “Statistical Analysis‌‌ of Networks”': (12h of lectures, 18H eq. TD,‌ and 12h of TD),‌ M2 Data Science and‌‌ Artificial Intelligence, Université Côte d’Azur (2025).
  - Teaching‌ practicals of the master‌ course “Statistical Inference”, (30H‌‌ TD), taught by Vincent Vandewalle, M1 Data‌ Science and Artificial Intelligence‌, Université Côte d’Azur‌‌ (2025).
- Josiane Zerubia gave 4h of lectures (6h‌ eq. TD), Master RISKS‌, Université Côte d’Azur,‌‌ France (2025). This course was given to Master‌ students.
10.2.2 Supervision
- Priscilla‌ Indira Osa did the‌‌ following supervision:
  - Co-supervision of a Master student in‌ 2025 for the Master‌ in Engineering for Natural‌‌ Risk Management, University of Genoa.
- Josiane Zerubia‌ supervised:
  - Two PhD students‌ (Louis Hauseux, and‌‌ Priscilla Indira Osa) within the AYANA team.‌
10.2.3 Juries
- Josiane Zerubia‌ participated to the following‌‌ jury:
  - Member of one CSI (Comité de Suivi‌ Individuel) PhD committee as‌ expert and another one‌‌ as student supervisor.
  - HDR‌ committee of Lamberto Dell'Elce at Université Côte d'Azur.‌
10.3 Popularization
- Louis Hauseux gave a talk at‌ 3IA seminar, October 2025.
- Josiane Zerubia participated‌ as a member of Inria/Terra Numerica/Femmes&Sciences to the‌ “
  10.3.1 Productions (articles, videos, podcasts, serious games,‌ ...)
  - Martina Pastorino and Josiane Zerubia are mentioned‌ as Women Role Models on Femmes&Sciences website.‌
  - Josiane Zerubia was interviewed by Agnès Bessières within‌ the national Inria community, December 2025.
  11 Scientific‌ production
  11.1 Major publications
  
  1articleJ.Jules‌ Mabon, M.Mathias Ortner and J.Josiane‌ Zerubia. Learning Point Processes and Convolutional Neural‌ Networks for Object Detection in Satellite Images.‌Remote Sensing166March 2024HAL DOI‌
  
  2articleM.Martina Pastorino, G.Gabriele‌ Moser, S. B.Sebastiano B. Serpico and‌ J.Josiane Zerubia. CRFNet: A Deep Convolutional‌ Network to Learn the Potentials of a CRF‌ for the Semantic Segmentation of Remote Sensing Images‌.IEEE Transactions on Geoscience and Remote Sensing‌2024, 1-19HALDOI
  11.2 Publications of‌ the year
  
  International journals
  
  3articleL.Louis‌ Hauseux, K.Konstantin Avrachenkov and J.Josiane‌ Zerubia. Generalization of Single-Linkage with Higher-Order Interactions‌.Applied Network ScienceNovember 2025. In‌ press. HAL DOI back to text
  
  4article‌M.Martina Pastorino, G.Gabriele Moser,‌ S. B.Sebastiano B. Serpico and J.Josiane‌ Zerubia. Probabilistic Graphical Models Meet Deep Learning‌ for Semantic Segmentation.IEEE Signal Processing Magazine‌December 2025. In press. HAL back to‌ text
  
  International peer-reviewed conferences
  
  5inproceedingsL.Louis‌ Hauseux, R.Raphaël Antoine, P.Philippe‌ Foucher, P.Pierre Charbonnier and J.Josiane‌ Zerubia. Généralisation du filtre de FRANGI pour‌ l'extraction des réseaux de fissures au sein d'images‌ d'un glissement de terrain. Comparaison avec une méthode‌ d'apprentissage profond.GRETSI 2025 - XXXe Colloque‌ Francophone de Traitement du Signal et des Images‌Strasbourg, FranceAugust 2025HAL back to text‌back to text back to text
  
  6inproceedings‌L.Louis Hauseux. How can we theoretically‌ measure the performance of density-based clustering algorithms?ACM‌ SIGMETRICS 2024 Student Research Competition524Venice,‌ ItalyMarch 2025, 19-20HAL DOI back‌ to text
  
  7inproceedingsP. I.Priscilla Indira‌ Osa, G.Gabriele Moser, S. B.‌Sebastiano B Serpico and J.Josiane Zerubia.‌ Joint Deep Missing Value Imputation and Clustering of‌ Satellite Image Time Series.EUSIPCO 2025 –‌ 33rd IEEE European Signal Processing ConferencePalermo, Italy‌September 2025HAL back to text
  
  8inproceedings‌P. I.Priscilla Indira Osa, R.Roozbeh‌ Rajabi, G.Gabriele Moser, S. B.‌Sebastiano B Serpico and J.Josiane Zerubia.‌ A Comparative Study of Unsupervised Winter/Summer Crop Classification‌ Methods for FPAR Image Time Series.IEEE‌ IGARSS 2025 – International Geoscience and Remote Sensing‌ SymposiumBrisbane, AustraliaAugust 2025HAL back to‌ text
  
  9inproceedingsM.Martina Pastorino, G.Gabriele Moser, S.‌ B.Sebastiano B Serpico‌ and J.Josiane Zerubia‌‌. CFC-MCRF: Multiresolution Fusion and Segmentation of Hyperspectral‌ and Panchromatic Remote Sensing‌ Images with Deep Learning‌‌ and CRFs.EUSIPCO 2025 – 33rd IEEE‌ European Signal Processing Conference‌Palermo, ItalySeptember 2025‌‌HAL back to text
  
  10inproceedingsM.Martina‌ Pastorino, G.Gabriele‌ Moser, S. B.‌‌Sebastiano B. Serpico and J.Josiane Zerubia.‌ Multiresolution Fusion and Classification‌ of Hyperspectral and Panchromatic‌‌ Remote Sensing Images.IEEE/CVF Winter Conference on‌ Applications of Computer Vision‌ Workshops (in WACV 2025)‌‌Tucson (AZ), United StatesFebruary 2025HAL
  11.3‌ Cited publications
  
  11article‌J. A.John A.‌‌ Hartigan. Consistency of Single Linkage for High-Density‌ Clusters.J. of‌ the Am. Stat. Ass.‌‌763741981, 388-394DOI back to‌ text
  
  12inproceedingsR.‌ J.Ricardo J. G.‌‌ B. Campello, D.Davoud Moulavi and J.‌Joerg Sander. Density-Based‌ clustering based on hierarchical‌‌ density estimates.Advances in Knowledge Discovery and‌ Data MiningBerlin, Heidelberg‌Springer2013, 160--172‌‌DOI back to text
  
  13inproceedingsH.Hugo‌ Duminil-Copin. Sixty years‌ of percolation.Proceedings‌‌ of the International Congress of Mathematicians, Rio de‌ JaneiroWorld Scientific2018‌, 2829--2856DOI back‌‌ to text
  
  14articleK.Kazimierz Florek,‌ J.Jan Łukaszewicz,‌ J.Julian Perkal,‌‌ H.Hugo Steinhaus and S.Stefan Zubrzycki.‌ Sur la liaison et‌ la division des points‌‌ d'un ensemble fini.Colloquium Mathematicum23-4‌1951, 282-285DOI‌back to text
  
  15‌‌inproceedingsA. F.Alejandro F. Frangi, W.‌ J.Wiro J. Niessen‌, K. L.Koen‌‌ L. Vincken and M. A.Max A. Viergever‌. Multiscale vessel enhancement‌ filtering.Medical Image‌‌ Computing and Computer-Assisted Intervention --- MICCAI'98Springer Berlin‌ Heidelberg1998DOI back‌ to text
  
  16article‌‌J. C.J. C. Gower and G. J.‌G. J. S. Ross‌. Minimum Spanning Trees‌‌ and Single Linkage Cluster Analysis.Journal of‌ the Royal Statistical Society.‌ Series C (Applied Statistics)‌‌1811969, 54--64DOI back to‌ text
  
  17articleW.‌Wenqi Guo, Z.‌‌Zheng Zhang, Y.Yu Meng, W.‌Weixiong Zhang, S.‌Shichen Gao and others‌‌. Deep Temporal Joint Clustering for Satellite Image‌ Time-Series Analysis.JSTARS‌182025, 1272-1287‌‌DOI back to text
  
  18inproceedingsL.Louis‌ Hauseux, K.Konstantin‌ Avrachenkov and J.Josiane‌‌ Zerubia. Hypergraphs, percolation, and hierarchical clustering.‌The 13th International Conference‌ on Complex Networks and‌‌ their ApplicationsStudies in Computational IntelligenceIstanbul, Turkey‌December 2024back to‌ text back to text‌‌
  
  19articleQ.Qianli Ma, C.Chuxin‌ Chen, S.Sen‌ Li and G. W.‌‌Garrison W. Cottrell. Learning Representations for Incomplete‌ Time Series Clustering.‌AAAI3510May‌‌ 2021, 8837-8846URL: https://ojs.aaai.org/index.php/AAAI/article/view/17070DOI back to‌ text back to text‌
  
  20inproceedingsL.Leland‌‌ McInnes and J.John‌ Healy. Accelerated Hierarchical Density Based Clustering.‌IEEE International Conference on Data Mining Workshops (ICDMW)‌2017, 33-42DOIback to text
  
  21‌articleM.M. Pastorino, G.G. Moser‌, F.F. Guerra, S. B.S.‌ B. Serpico and J.J. Zerubia. Probabilistic‌ Fusion Framework Based on Fully Convolutional Networks and‌ Graphical Models for Burnt Area Detection from Multiresolution‌ Satellite and UAV Imagery.IEEE Transactions on‌ Geoscience and Remote SensingAccepted for publication2024‌back to text
  
  22articleM.M. Pastorino‌, G.G. Moser, S. B.S.‌ B. Serpico and J.J. Zerubia. Cross-Modal‌ Fusion and Classifi- cation of Hyperspectral and Panchromatic‌ Remote Sensing Images with Deep Learning and Multiscale‌ CRFs.IEEE Transactions on Geoscience and Remote‌ SensingIn revision2025back to text
  
  23‌bookM.Mathew Penrose. Random Geometric Graphs‌.5Oxford University Press2003DOI back‌ to text
  
  24inproceedingsJ.Junyuan Xie,‌ R.Ross Girshick and A.Ali Farhadi.‌ Unsupervised Deep Embedding for Clustering Analysis.ICML‌48Proceedings of Machine Learning ResearchNew York,‌ New York, USAPMLR20--22 Jun 2016,‌ 478--487URL: https://proceedings.mlr.press/v48/xieb16.htmlback to text back to‌ text
  
  25miscS.Shanglian Zhou, C.‌Carlos Canchila and W.Wei Song. Fused‌ Image dataset for convolutional neural Network-based crack Detection‌ (FIND).March 2022DOI back to text‌

AYANA - 2025

AYANA - 2025

2025Activity reportTeam﻿﻿﻿‌AYANA

Keywords﻿​﻿﻿

Computer Science and Digital​‌﻿﻿ Science

Other Research Topics​‌﻿﻿ and Application Domains

1 Team members,​​﻿﻿ visitors, external collaborators

Research​​​‌ Scientist

PhD Students​​​‌

Administrative Assistant​​​‌

External Collaborators

2 Overall​‌﻿﻿ objectives

3 Research​​﻿﻿ program

3.1 FAULTS R​​​‌ GEMS: Properties of faults,﻿​﻿﻿ a key to realistic​‌﻿﻿ generic earthquake modeling and​​﻿﻿ hazard simulation

3.2 Probabilistic﻿﻿﻿‌ models on graphs and﻿‌​‌ machine learning in remote﻿​​﻿ sensing applied to natural​​​‌ disaster response

3.3 Marked﻿﻿﻿‌ point process models for﻿‌​‌ object detection and tracking﻿​​﻿ in temporal series of​​​‌ high resolution images

4 Application domains​‌﻿﻿

5 Highlights﻿​﻿﻿ of the year

5.1​‌﻿﻿ Awards

6​​​‌ Latest software developments, platforms,﻿​﻿﻿ open data

6.1 Latest​‌﻿﻿ software developments

6.1.1 MPP​​﻿﻿ & CNN for object​​​‌ detection in remotely sensed﻿​﻿﻿ images

6.1.2 FCN and﻿﻿﻿‌ Fully Connected NN for﻿‌​‌ Remote Sensing Image Classification﻿​​﻿

6.1.3 Stabilizer for﻿﻿﻿‌ Satellite Videos

6.1.4﻿​​﻿ GLMB filter with History-based​​​‌ Birth

6.1.5 Automatic fault﻿﻿﻿‌ mapping using CNN

6.1.6 RFS-filters for﻿‌​‌ Satellite Videos

6.1.7​​​‌ AYANet

6.1.8 GabFormer​​​‌

6.1.9 PP-EBM​​﻿﻿

6.1.10 HGP-K2

6.1.11 CFC-CRF

6.1.12 BAS-UAV

7​​​‌ New results

7.1 Probabilistic﻿​﻿﻿ graphical models and convolutional​‌﻿﻿ networks for the semantic​​﻿﻿ segmentation

7.2 A Probabilistic Fusion​​​‌ Framework for Burnt zones﻿﻿﻿‌ mapping from multimodal satellite﻿‌​‌ and UAV imagery

7.3 Multiresolution Fusion and﻿‌​‌ Segmentation of Hyperspectral and﻿​​﻿ Panchromatic Remote Sensing Images​​​‌ with deep learning and﻿﻿﻿‌ probabilistic graphical models

7.4 Joint Deep-learning-based﻿​﻿﻿ Gap-filling and Clustering of​‌﻿﻿ Satellite Image Time Series​​﻿﻿

7.5 Generalization﻿‌​‌ of Single-Linkage: Hypergraphs and﻿​​﻿ Percolation Analysis

7.6 Generalized​​﻿﻿ Frangi Filter: Multimodal Fusion​​​‌ for Crack Extraction

7.7 3D﻿‌​‌ LiDAR Instance Segmentation with﻿​​﻿ Geometric Priors: The HGP-Clusterer​​​‌

8​​﻿﻿ Bilateral contracts and grants​​​‌ with industry

8.1 Bilateral﻿​﻿﻿ contracts with industry

8.1.1​‌﻿﻿ LiChIE contract with Airbus​​﻿﻿ Defense and Space funded​​​‌ by Bpifrance

9 Partnerships​​​‌ and cooperations

9.1 International﻿​﻿﻿ research visitors

9.1.1 Visits​‌﻿﻿ to international teams

Research​​﻿﻿ stays abroad

Josiane Zerubia​​​‌

9.2﻿​﻿﻿ National initiatives

ANR FAULTS​‌﻿﻿ R GEMS

9.3 Prospecting​​​‌ industrial partners

10 Dissemination​​​‌

10.1 Promoting​​​‌ scientific activities

10.1.1 Scientific﻿​﻿﻿ events: selection

Member of​‌﻿﻿ the conference program committees​​﻿﻿

Reviewer

10.1.2 Journal

Member of​​​‌ the editorial boards

Reviewer -﻿﻿﻿‌ reviewing activities

10.1.3 Invited talks﻿​​﻿

10.1.4﻿‌​‌ Leadership within the scientific﻿​​﻿ community

10.1.5 Scientific﻿​​﻿ expertise

10.1.6 Research administration

10.2﻿﻿﻿‌ Teaching - Supervision -﻿‌​‌ Juries - Educational and﻿​​﻿ pedagogical outreach

10.2.1 Supervision​​​‌

10.2.2 Supervision

10.2.3 Juries

10.3 Popularization

10.3.1 Productions (articles,﻿​﻿﻿ videos, podcasts, serious games,​‌﻿﻿ ...)

11 Scientific​‌﻿﻿ production

11.1 Major publications​​﻿﻿

11.2 Publications of​‌﻿﻿ the year

International journals​​﻿﻿

International peer-reviewed conferences​​﻿﻿

11.3​​​‌ Cited publications

2025Activity reportTeam‌AYANA

Keywords

Computer Science and Digital‌ Science

Other Research Topics‌ and Application Domains

1 Team members, visitors, external collaborators

Research‌ Scientist

PhD Students‌

Administrative Assistant‌

2 Overall‌ objectives

3 Research program

3.1 FAULTS R‌ GEMS: Properties of faults, a key to realistic‌ generic earthquake modeling and hazard simulation

3.2 Probabilistic‌ models on graphs and‌‌ machine learning in remote sensing applied to natural‌ disaster response

3.3 Marked‌ point process models for‌‌ object detection and tracking in temporal series of‌ high resolution images

4 Application domains‌

5 Highlights of the year

5.1‌ Awards

6‌ Latest software developments, platforms, open data

6.1 Latest‌ software developments

6.1.1 MPP & CNN for object‌ detection in remotely sensed images

6.1.2 FCN and‌ Fully Connected NN for‌‌ Remote Sensing Image Classification

6.1.3 Stabilizer for‌ Satellite Videos

6.1.4 GLMB filter with History-based‌ Birth

6.1.5 Automatic fault‌ mapping using CNN

6.1.6 RFS-filters for‌‌ Satellite Videos

6.1.7‌ AYANet

6.1.8 GabFormer‌

6.1.9 PP-EBM

7‌ New results

7.1 Probabilistic graphical models and convolutional‌ networks for the semantic segmentation

7.2 A Probabilistic Fusion‌ Framework for Burnt zones‌ mapping from multimodal satellite‌‌ and UAV imagery

7.3 Multiresolution Fusion and‌‌ Segmentation of Hyperspectral and Panchromatic Remote Sensing Images‌ with deep learning and‌ probabilistic graphical models

7.4 Joint Deep-learning-based Gap-filling and Clustering of‌ Satellite Image Time Series

7.5 Generalization‌‌ of Single-Linkage: Hypergraphs and Percolation Analysis

7.6 Generalized Frangi Filter: Multimodal Fusion‌ for Crack Extraction

7.7 3D‌‌ LiDAR Instance Segmentation with Geometric Priors: The HGP-Clusterer‌

8 Bilateral contracts and grants‌ with industry

8.1 Bilateral contracts with industry

8.1.1‌ LiChIE contract with Airbus Defense and Space funded‌ by Bpifrance

9 Partnerships‌ and cooperations

9.1 International research visitors

9.1.1 Visits‌ to international teams

Research stays abroad

Josiane Zerubia‌

9.2 National initiatives

ANR FAULTS‌ R GEMS

9.3 Prospecting‌ industrial partners

10 Dissemination‌

10.1 Promoting‌ scientific activities

10.1.1 Scientific events: selection

Member of‌ the conference program committees

Member of‌ the editorial boards

Reviewer -‌ reviewing activities

10.1.3 Invited talks

10.1.4‌‌ Leadership within the scientific community

10.1.5 Scientific expertise

10.2‌ Teaching - Supervision -‌‌ Juries - Educational and pedagogical outreach

10.2.1 Supervision‌

10.3.1 Productions (articles, videos, podcasts, serious games,‌ ...)

11 Scientific‌ production

11.1 Major publications

11.2 Publications of‌ the year

International journals

International peer-reviewed conferences

11.3‌ Cited publications