Bandeau traitement de données

RootSystemTracker: automatic methods for spatio-temporal reconstruction of root architecture using AI for agroecological system design

PhD research project awarded in the 2024 call for projects.

  • Title: "RootSystemTracker: automatic methods for spatio-temporal reconstruction of root architecture using AI for agroecological system design"; Inria & Cirad.
    • PhD student: Gandeel Loaï.
    • Affiliated unit: University of Montpellier & CNRS, UMR 5506 - LIRMM.
    • Co-supervision:
      • Reza Akbarinia (Inria).
      • Romain Fernandez (Cirad).
    • Doctoral school: I2S – Information Structures Systèmes.
    • Project duration: 2024 – 2027.

Project summary:

Current automatic root phenotyping methods are limited by occlusions, the complexity of root structures, and variability in observation conditions, making dynamic and comprehensive analysis of root growth challenging in field or mixed crop conditions. These approaches only partially solve the reconstruction problem in simple, controlled cases, without guaranteeing a globally optimal solution. A major challenge in studying root architecture is the lack of visualization of root system structures over time in their natural environment. To overcome these challenges, RootSystemTracker will leverage recent advances in spatio-temporal analysis of root architecture (Fernandez et al., 2022), combining global graph algorithms with deep learning and data mining techniques to analyze root trajectories. This project will transfer these advances from the laboratory to the field using training data provided by an international consortium and shared through a data challenge. The data will be produced by phenotyping devices allowing the observation of roots in the form of time series: rhizotrons, rhizotubes, rhizoscopes, and optical scanners. The main objectives of the project include developing open-source software for the annotation and spatio-temporal reconstruction of root architectures, training models for domain adaptation and semantic segmentation, and improving analysis models with data mining. RootSystemTracker will strengthen international cooperation and data sharing between agronomists and data science experts. The project will encourage resource accessibility, allowing both Northern and Southern countries to benefit from methodological advances. It will improve understanding of root development dynamics in response to environmental stresses and species mixtures, enabling the modeling and selection of varieties adapted to agroecology.

 

GIF root tracker
Suivi d'un système racinaire en croissance et reconstruction de son architecture (Arabidopsis).