🧑 Mihai Ivanovici (UTBV); Josiane Mothe (UT); Andreea Nițu (UTBV); Matei Debu (UTBV)
📅 Mon. 17 – Fri. 21 March 2025
From Monday 17 to Friday 21 March 2025 the University of Toulouse hosted a delegation from Transylvania University of Brașov as part of the AI4AGRI European precision agriculture project. The visit included research presentations, discussions on project development, and strategic meetings to enhance European collaboration.
Monday
AI4AGRI 6.0 Preparation
Members of the University of Toulouse and the visiting delegation from Transylvania University of Brașov initiated discussions on the development of AI4AGRI 6.0.
Kamal Marandskiy’s PhD Defense Rehearsal
Kamal Marandskiy presented a rehearsal of his PhD defense at IRIT, focusing on remote sensing data analysis for Earth observation applications. His research addresses:
- Soil roughness estimation using multispectral data from ground-level images, UAV captures, and Sentinel-1 SAR.
- Hyperspectral image compression and visualization, where he introduced a novel lossy compression technique based on Fourier Transform and exponential feature extraction, achieving significant data reduction while maintaining information integrity.
Marandskiy’s work includes published research on hyperspectral segmentation, data fusion for potato field identification, and machine learning-based classification of Sentinel-1 backscattering coefficients. His contributions to soil roughness estimation and hyperspectral image visualization are particularly relevant for precision agriculture applications.
Tuesday
Prof. Mihai Ivanovici and Prof. Josiane Mothe met with the Europe Unit of the University of Toulouse, presenting strategies for improving the Romanian Excellence Center on AI for Agriculture under the WIDERA funding call. The discussions were complemented by meetings with the IRIT partnerships department and the European projects service at UT2J. These sessions provided valuable insights for the Romanian delegation on structuring EU research proposals and leveraging institutional support mechanisms.
Wednesday
A full day of workshops facilitated knowledge exchange between researchers from Brașov Transylvania University and the University of Toulouse, covering remote sensing applications in precision agriculture.
Mihai Ivanovici: Spectral-Temporal Analysis of Agricultural Crops Using Sentinel-2 Data
Prof. Ivanovici presented findings on NDVI temporal patterns for crop identification. His analysis, conducted in collaboration with the National Institute of Research and Development for Potato and Sugar Beet Braşov, examined inter-annual variations in NDVI curves across different crops. His research challenges the assumption that NDVI evolution is a consistently reliable crop identifier, highlighting environmental influences and seeding frequency as critical factors.
Matei Debu: Creation of the DACIA-5 Dataset
MSc student Matei Debu introduced the DACIA-5 dataset, a comprehensive Sentinel-1 and Sentinel-2 time-series dataset for crop identification in the Brașov region (2020-2024). His presentation detailed:
- Data acquisition and preprocessing steps.
- File naming conventions and ground-truth labeling.
- Applications in crop identification and early-season classification using ResNet18 and Random Forest models.
Andreea Nitu: Vegetation Indices for Precision Agriculture
PhD student Andreea Nitu explored the role of vegetation indices in precision agriculture, presenting:
- A UAV-acquired hyperspectral dataset.
- A comparative analysis of raw hyperspectral pixel values versus vegetation indices for crop classification.
- The potential of vegetation indices for dimensionality reduction prior to machine learning applications.
Mohammad El Sakka: Ongoing research in Feature Map Denoising and paper review
Mohammad El Sakka presented his ongoing research on Feature Map Denoising for Sparse Image Segmentation where he introduced a novel method to remove structured noise from deep learning feature maps, improving segmentation performance in UAV-based agricultural datasets.
In addition, El Sakka shared his review of the paper SpectralGPT: Spectral Remote Sensing Foundation Model, which is a potentially relevant pre-trained model for AI4AGRI members, who could make use of it for Sentinel-2 Earth Observation applications.
Hugo Carlesso: Application of Curriculum Learning in Hyperspectral Image Classification
Hugo Carlesso presented his ongoing research on the application of Curriculum Learning in combination with Jigsaw and Masked Image Modelling for HYPSO-1 hyperspectral image classification.
Serge Molina: Dense Passage Retrieval in Low-Resource Contexts
Serge Molina presented the use of a new dataset for training dense passage retrieval models in the low-resource context of Amharic information retrieval.
Pape Ibrahima THIAM: Improving Named Entity Recognition in Specialized Domains
Pape Ibrahima THIAM presented his work recently submitted on enhancing named entity recognition in specialized domains, approach leveraging thematic segmentation and passage retrieval.
Friday
Mihai Ivanovici, Olivier Teste, and Josiane Mothe discussed the evaluation of the AI4AGRI project. Josiane Mothe provided guidance on how to consolidate and expand the capacities of AI4Agri. Mihai, Josiane and Olivier worked on the proposal in a brainstorming and writing session. Olivier Teste, who has participated in numerous evaluations of CNRS laboratories in France as part of the High Council for the Evaluation of Research and Higher Education (HCERES), provided insights into best practices for research assessment. Guidance was given on the structure and content of the project evaluation report, helping align AI4AGRI’s self-assessment with recognized European standards.
Among advices provided by Université de Toulouse are:
- the importance of selecting the journals and conferences where to publish
- the importance of the administration services and research supports
Students feedback
From March 17 to March 21, we, Andreea NITU and Matei DEBU, research assistants, accompanied Professor Mihai IVANOVICI on a project-related visit to Université de Toulouse (ex. Université Toulouse 3) as well as affiliated entities, the Institute of Research in Informatics of Toulouse (IRIT) is affiliated to. This visit was part of an ongoing partnership aimed at strengthening research collaboration between our institutions. As junior researchers, our primary objective was to meet the Toulouse team and establish stronger professional connections with the institute’s members.
Now, at the end of our visit we can summarize our experience as being very productive, informative and useful for our future academic activity. To enumerate some of the activities, we have:
- Met prof. Josiane Mothe. She provided guidance and very useful feedback on how to improve the presentation of our research work.
- Assisted the discussion between prof. Mihai Ivanovici, prof. Josiane Mothe and other collaborators about the future project that they prepare which provided us with valuable insights into project management.
- Met prof. Josiane PhD students and some research assistants that participated in an workshop where we all presented our current work and exchanged some ideas on cutting-edge technologies, articles and some feedback to each others presentations.
- We deepened our knowledge about article writing, figure generating and state-of-the-art research methodologies with the guidance of Phd student Mohammad El-Sakka.
- We walked around the Université Toulouse 3 Paul Sabatier campus and had some social activities with colleagues from IRIT.
- Although we were in France, we stayed connected with our home institution in Brașov attending to the remote preliminary Phd defense of our colleague, Kamal Marandskiy.
To conclude, we think that the experience was a success: we enrich our knowledge, gained firsthand experience of successful project administration and straighten our professional relationships with the AI4AGRI colleagues from Toulouse.