Work

FarrowSight: An intelligent system for early-stage piglet growth performance prediction in farrowing stables

Year: 2026

Type: article

Abstract: Accurate prediction of pre-weaning piglet growth curves is essential for forecasting weaning weight, a pivotal indicator of piglets’ future development and genetic breeding potential. Traditionally, recording growth curves relies on daily manual weighing, which is labor-intensive, induces stress in piglets, and is unsuitable for continuous monitoring. To address these limitations, it is imperative to develop a system that enables non-contact individual weight monitoring and early-stage prediction of pre-weaning growth curves. This study introduces FarrowSight, an intelligent system integrated with a Red Green Blue-Depth (RGB-D) camera, designed to identify freely moving piglets non-contact and estimate each piglet’s instantaneous weight in farrowing stables. Concurrently, the AutoGluon-based Iterative Network (AG-IterNet) algorithm was developed to enable precise monitoring of individual piglet time-series growth dynamics based on instantaneous weight measurement, achieving the prediction of pre-weaning growth curves as early as possible. FarrowSight exhibited exceptional predictive accuracy for pre-weaning growth curves using only the first week of weight data, achieving a coefficient of determination (R 2 ) of 0.827 (95 % confidence interval (CI): 0.816, 0.838) and a Mean Absolute Percentage Error (MAPE) of 10.833 % (95 % CI: 10.526 %, 11.139 %). Moreover, prediction performance demonstrated progressive enhancement with the incorporation of additional early-stage weight measurements, effectively advancing the assessment timeline from traditional 3–4 week weaning weights to the critical first post-birth week. This innovation holds significant potential for optimizing feeding management and selecting superior individuals within the swine industry. • FarrowSight is a non-contact growth curve prediction system for free-moving piglets. • AG-IterNet predicts pre-weaning growth curves (R 2 =0.827) with only 1-week data. • Early growth curve prediction advances genetic selection from 3 to 4 weeks to 1 week. • Automated growth analytics replace labor-heavy manual weighing, boosting efficiency.

Source: Computers in Industry

DOI: 10.1016/j.compind.2025.104433