[Defended thesis] Maëva Durand

[Defended thesis] Maëva Durand: Automated on real-time welfare and health assessment of gestating sows using heterogeneous data

Maëva defended her PhD on 23 October 2023 @Institut Agro

Automated on real-time welfare and health assessment of gestating sows using heterogeneous data

My name is Maëva Durand, and I'm doing my thesis with two research units: Inrae's UMR Pegase, based in Saint-Gilles, and Inria's Lacodam, located in Rennes.

After graduating with an Agronomy Engineering degree in Animal Production Science and Engineering, I wanted to continue deepening my knowledge of pig production by pursuing a PhD. First of all, this thesis will enable me to perfect my knowledge of scientific methodology: design and implementation of protocols, monitoring of experiments, sorting and analysis of data, writing scientific articles in English, etc. These skills will be useful for continuing my studies in the future. These skills will help me to continue working in the swine research sector.

My topic concerns the real-time integration of welfare and health indicators for pregnant sows into nutritional models, used in particular for precision feeding. These indicators will be used to trigger alerts and actions within an innovative breeding system. This thesis will be based on complementary approaches mobilizing two main scientific fields: animal sciences, more specifically sow behavior and health, and computer science, in particular artificial intelligence, to automate the extraction of knowledge from data and their representation for the purposes of supervision and decision support.

This thesis will provide me with further knowledge in the field of Machine Learning and animal nutrition modeling, two areas which I feel are essential for improving conditions on pig farms. The ambition of the project is to produce an operational decision-making tool for improving the health and well-being of sows, which is both flexible and robust enough to adapt to the variability of the amount of information available on the farm.

  • Starting date: 1st October 2020
  • University: AgroCampus Ouest
  • PhD school: Ecole Doctorale Écologie, Géosciences, Agronomie, Alimentation (EGAAL)
  • Scientfic field: Agriculture, Animal sciences
  • Thesis management: Jean-Yves Dourmand, Pegase, Inrae
  • Thesis supervisors: Charlotte Gaillard, Pegase, Inrae et Christine Largouët, Irisa, Inria
  • Funding: #DigitAg – Inrae
  • #DigitAg : Cofunded thesis – Axe 5 : Fouille de données, analyse de données, extraction de connaissances, Axe 2 : Innovations en agriculture numérique – Challenge 2 : Le phénotypage rapide, Challenge 4 : Des productions animales durables,

Keywords: Behavior, well-being, precision breeding, technology, sows

Résumé : Evaluating sows’ behaviour allows the early detection of health or welfare problems and the quantification of their physical activity, which is major factor affecting their energy requirement. In practice, the continuous observation of each animal by the breeder is impossible and only intermittent observations are performed, often only once a day. New technologies have been developed like the accelerometers measuring the activity of each animal and its behaviour as well as software analysing animals’ vocalizations. Video analysis also offers the possibility to study social interactions between animals. Feeding and drinking behaviours may also be collected by feed and water dispensers. These behavioural data coupled with production data (feed intake, body weight and backfat thickness) should allow real-time estimation of the welfare of each animal, estimated through different criteria, and anticipate the occurrence of health problems in a non-invasive manner. The innovative issue is to describe the behavioural models using timed automata. The learning of timed automata is a rather new area of research. Their interest lies in time representation, the efficiency of exploration (using model-checking techniques) and the explainability of the models for predictive and monitoring tasks. The objectives of this thesis are therefore, in a first step, (i) to collect behavioural and production data in different situations (non-stressful vs. stressful), (ii) to learn and model the behaviours with timed automata describing gestating sows and the relationships between them in different situations (iii) to set up alerts and actions system to improve welfare, and (iv) to test in the experimental farm the capacity of this system to improve welfare and health.

Contact : durand.maeva22 [AT] gmail.com​ - Tél: 06.89.06.76.26

Social network: LinkedIn