📅 Tue. 24 – Sat. 28 September 2024
Șerban will participate to the UT3 team weekly research meeting as well as several PhD students’ presentations in relation to AI4Agri research topics. He will present him self some of his work on Image analysis (features, segmentation) for applications in agriculture. He will meet the different AI4Agri UT3 members and they will present their role. He will be trained on communication tools used by the UT3 team for producing web articles and the newsletter.
Visit Report
Wednesday 25/09/2024 (Day 1)
09:00 | SIG team weekly meeting
Șerban OPRIȘESCU stay in Toulouse began with the SIG team weekly meeting where the SIG team members introduced Șerban OPRIȘESCU to their research topics and shared their research progress.
11:00 | Presentation: Image analysis (features, segmentation) for applications in agriculture
Following the team meeting, Șerban OPRIȘESCU proceeded to present a research work that has been conducted on the use of the Shannon-Weaver biodiversity index for grassland quality estimation, taking advantage of hyper-spectral satellite images.
An outcome of this presentation has been the organization of a presentation by Pascal DEYRE during the second day on the production and distribution of datasets following F.A.I.R. principles.
12:00 | Presentation: Bayesian Sparse Model for Complex-Valued Magnetic Resonance Spectroscopy Restoration
This presentation focused on efficiently quantifying metabolite concentrations potentially involved in glioblastoma (GBM) relapse, a highly aggressive tumor with poor outcomes. The study expanded the target database from four to 18 metabolites, aiming to enhance biomarker discovery.
The MR spectral reconstruction model was based on representing each spectrum as a linear combination of metabolite signals stored in a dictionary, with each signal weighted by its concentration. A novel Bayesian approach for MRS restoration was introduced, which accounted for errors in the observation linear operator. This method employed sparse regularization, incorporating both the real and imaginary parts of the spectra to address the complex nature of MRS data.
A Bernoulli-exponential prior was used to enforce sparsity, allowing for the efficient selection of relevant metabolites. Inference was conducted using a Gibbs sampling strategy within a Markov Chain Monte Carlo (MCMC) framework.
The results were coherent with state-of-the-art methods while enforcing the sparsity constraint, validating the effectiveness of the approach.
Mid-day break: Restaurant Brasserie Ponsan
Starter
Truite marinée aux baies roses
Feuilleté poire et roquefort
Main course
Tataki de saumon
Émincé de veau au miel et épices
Desert
Mousse au chocolat
15:00 | Presentation: Denoising Feature Maps for Enhanced Weed Detection in Sparse Vegetation
During this presentation Mohammad El Sakka shared his results on a yet to be published research work where a denoising layer is introduced between the convolution layers of a U-net fully convolutional neural network trained on crop/weed/soil classification.
16:00 | Brainstorming: AI in Agriculture
After Mohammad El Sakka’s presentation, the researchers participated in a brainstorming session on the topic of AI in Agriculture, Josiane Mothe led the session and collected the ideas which will be used for defining new research directions in the AI4AGRI project.
Friday 27/09/2024 (Day 2)
10:00 | Presentation: dataset distribution with F.A.I.R. principles
Following the request from Șerban OPRIȘESCU on how to distribute a dataset produced in the context of the AI4AGRI project, Pascal Dayre from SO-DRIIHM project introduced the team to F.A.I.R. principles.
During his presentation, he answered questions such as:
– Who can host datasets, under what conditions, and with what services?
– Where can datasets be stored?
– How to document a dataset?
– Can datasets be hosted privately while adhering to FAIR principles?
– Ethical and legal aspects
Mid-day break: Restaurant L’esplanade
Starter
Tartelette de poireaux façon tatin
Salade de carottes à la marocaine
Main course
Tourte de pintade sauce diable et salade
Pdt roseval gratinées à la raclette
Blanquette de veau, purée de pomme de terre
Desert
Tarte chocolat
Cheese cake
Tarte tatin
Poire au sirop
Salade d’oranges
Crumble
13:00 Website administration knowledge sharing
For the last part of Șerban OPRIȘESCU visit, Serge Molina who is currently the co-author of the AI4AGRI newsletter, introduced Șerban to the process of producing and distributing the newsletter as well as how to access the website analytics for studying the audience and reach of the AI4AGRI website.