Automated computer vision analysis of pig appeasing pheromone effects on social behaviour and welfare in group-housed sows

In the EU, the reproductive cycle of sows commonly includes prolonged periods of individual housing during the farrowing and insemination phase, followed by the mandatory group housing during the gestation phase. The regrouping of previously familiar and partly unfamiliar sows at the start of gestation triggers agonistic interactions for (re‑)establishing a social hierarchy. This period of social unrest is associated with stress and a high incidence of lesions and lameness, posing welfare and productivity challenges. The aim of this study was to investigate whether social unrest and its negative consequences could be alleviated by the application of a Pig Appeasing Pheromone (PAP) during the first week of the gestation period. In a double‑blind experiment, 8 groups of sows (16–20 per group) were randomly allocated to a PAP (9 diffuser blocks containing 2% SecurePig® distributed across the pen from 3d before until 7d after mixing) or placebo treatment (identical placement of diffuser blocks lacking the active PAP component) at the start of the gestation period. Skin lesions obtained after mixing were scored on d2, d7, d9 and d17. Four cameras per pen were used to video‑record the uniquely marked sows continuously from 8‑20h for 17d after mixing. An in‑housed developed and validated automated computer‑vision‑based system was used to enable high‑resolution, continuous tracking and analysis of individual and dyadic behaviour. Zero‑shot‑learning‑based segmentation, identification, and tracking enabled quantification of movement and agonistic interactions. Our AniDomNet tool was fitted separately to sequential fleeing and displacement interactions to infer social dominance matrices and directed acyclic graphs. Effects of treatment were tested using linear mixed models. PAP was associated with reduced overall movement (p = 0.002), fighting (p = 0.011), fleeing (p < 0.001), displacements (p < 0.001), and skin lesion scores (p = 0.046). The largest reduction occurred between d2 and d7 post‑grouping, following the initial peak in agonistic behaviour immediately after mixing. These effects were sustained following PAP diffuser removal at d7 without signs of a rebound effect. The AniDomNet hierarchies based on fleeing vs displacements showed 65% agreement, and indicated a broadly linear dominance structure with some unsettled dyads. The effects of PAP on hierarchy structure seemed marginal: PAP groups showed slightly less linear hierarchies (using Landau h) and less settlement at lower ranks. High‑ranked animals had more strongly preferred resting places, were on average further away from the feeding station, moved less (especially in control groups), were closer to each other during resting, and occupied areas less equally as compared to lower‑ranked sows. This study demonstrates that PAP administration can reduce social stress during early gestation grouping, and that integrating zero‑shot learning with continuous tracking provides a scalable, observer‑independent approach for advancing ethological research in group‑housed farm animals.