Phenomics

Continuous computer vision monitoring of sow posture linked to litter size in free-farrowing pens

---

Abstract

Growing political and public concern about sow welfare has driven interest in alternatives to conventional farrowing crates. These systems substantially restrict sow movement during farrowing and lactation, limiting the expression of natural behaviors. Free-farrowing pens offer greater freedom of movement and the opportunity to perform species-specific maternal behaviors but often increase piglet mortality from crushing, and complicate handling due to increased sow activity and aggression. Genetic selection targeting maternal behavior may help mitigate these challenges, but it requires accurate and continuous monitoring technologies. Recent advances in computer vision enable automated, non-invasive detection of sow posture and activity, providing novel opportunities to quantify maternal behavior over key reproductive periods. This study aimed to establish a proof of concept for continuous monitoring of sow postural behavior in a commercial free-farrowing environment and to investigate associations between maternal posture and reproductive performance. Video data were collected from 62 sows housed in free-farrowing pens in a Topigs Norsvin nucleus herd in Canada. A YOLOv8-based computer vision segmentation model was trained using 18,403 images to classify sow postures into six categories: lateral lying, sternal lying, semi-lateral lying, standing, sitting, and kneeling. Videos from two days pre-partum to two days post-partum were continuously analyzed at one frame per second. These postural data were aggregated per day relative to farrowing as average percentage of time spent in each posture and total number of posture transitions. Total number born (TNB) was used as the reproductive performance trait, with sows categorized as either high TNB (≥17 piglets) or low TNB (≤16 piglets). A linear mixed-effects model was fitted for each trait, including herd-year-season, parity, and the interaction between TNB class and day relative to farrowing as fixed effects, and sow as a random effect to account for repeated measures. Across all sows, lateral lying and standing decreased and sternal and semi-lateral lying decreased from pre- to post-partum. On the day prior to farrowing, low TNB sows spent less time lying (p < 0.001) and more time standing (p < 0.001) and kneeling (p < 0.01) compared to high TNB sows. On the day of farrowing, high TNB sows spent more time sitting (p < 0.001). Low TNB sows exhibited more postural transitions the day before farrowing (p < 0.001), but fewer transitions on the day of farrowing (p = 0.002) compared to high TNB. These findings demonstrate that maternal postural behavior around farrowing is associated with litter size, with the greatest differences occurring during the period of nest-building. The preliminary results suggest that sows producing larger litters exhibited reduced nest-building behavior. Future research will incorporate additional behavioral metrics and piglet survival to further elucidate the relationship between maternal behavior, genetic background, and reproductive success in free-farrowing systems.

Keywords: 2026

How to Cite:

van der Zande, L., Lei, V. & Grindflek, E., (2026) “Continuous computer vision monitoring of sow posture linked to litter size in free-farrowing pens”, World Congress on Genetics Applied to Livestock Production Digital Archive 2026(1): 2283084. doi: https://doi.org/10.31274/wcgalp.23478

Rights: 1