Abstract
Broiler breeder mating behaviour consists of rapidly performed sub-actions that determine the success of sperm transfer critical for egg fertility. It includes rooster mounting the hen followed by coordinated tail movements that enable cloacal contact. This study aimed to develop a vision-based deep learning (DL) model for automatically detecting the mounting phase, the period between mounting and decoupling during mating, and to evaluate its features over time. An experiment was conducted with four pens each containing 10 hens (Cobb 500F) and 1 rooster (Cobb MX strains) for four and a half months. Two cameras were stationed at each pen continuously recording broiler breeders daily. The DL model, YOLOv8, was fine-tuned and tested to detect two behavioural classes, mounting and non-mounting, with a dataset consisting of 2420 instances and further validated with independent dataset; the final model was applied on the experimental dataset to detect daily mounting phase events. Additionally, the welfare indicators footpad dermatitis (FPD) and gait score (GS), along with live weight, were periodically evaluated in broiler breeders and maintained within the Cobb recommended limits. Finally, a linear mixed model (LMM) was applied to evaluate individual and covariate effects of age, welfare and weight variables on the mounting phase results. The final model achieved 91.0 % and 90.0 % test accuracies in detecting mounting and non-mounting behaviours by roosters, respectively. The model detection results revealed an average 13.5 ± 4.2 mounts per rooster per day, with each one lasting 5.3 ± 0.8 seconds and happening with 70.6 ± 28.9 minutes inter-mounting interval. On average, 68.2 % of mounting phases were carried out with observable rooster and hen tail movements. Mountings that entailed tail movements lasted longer than the ones with none. The peak mounting activity was during the 16:00-20:00 followed by the 07:00-09:00 hours. Heavier roosters (p < 0.0001) and hens with poorer gait conditions (p = 0.0179) were less likely to engage in or sustain during mating. These findings may enhance understanding of the mating temporal dynamics and facilitate egg fertility analyses.