L'intelligence artificielle pour caractériser le comportement des chèvres laitières

International audience

Английский. In a context of growing societal demand for animal welfare focuses on the animal’s feelings and forced rules to change, despite incomplete knowledge [1]. Moreover, welfare is a complex concept, as with robustness and resilience and is not directly measurable [2]. Animal behavior is a key criterion to evaluate welfare. However, behavior data collection is costly and time-consuming, in particular because of its dynamic which requires following animal’s activities over a period of time [3]. The question addressed in this project was the impact of changing the number of feed distributions on goat behavior in term of time-budget. In order to answer this question, INRAE’s MoSAR research unit conducted a trial part of the “MaxForGoat” project in 2022 at the Grignon goat farm. Two feeding strategies (two or three meals a day) and two different feeds (conventionak mixed control or corn silage based) had been tested on four homogeneous groups of eight goats over three periods of two weeks. To follow goats behavior, one camera was located above each group of goats (four cameras). Moreover, in each group, four goats were equipped with accelerometers to measure acceleration in three axes with a 5hz frequency. My project, conducted as part of « la recherche et moi », aimed annotating videos using Boris© software (Behavioral Observation Research Interactive Software) start and end times for the four behaviors of interest (“ruminating”, “feeding”, “lying” and “standing”) of sixteen goats over three two hours periods. To characterize the behaviors related to the accelerometer data I used the machine learning model Modbehav developed as part of Sarah Mauny’s thesis and in order to evaluate the results I calculated AUC (Area Under the Curve) scores between model results and annotation. These scores are as follows : 0,58 for standing, 0,64 for lying 0,56 for rumination and 0,63 for feeding. Finally, I was able to see if as I expected, goats in groups receiving three meals a day expressed more feeding and rumination behaviors than goats in groups receiving two meals a day.