Comparison of machine learning algorithms for estimation of pig body weights from on-animal and digital image measurements

Wang, Zhuoyi
Journal Title
Journal ISSN
Volume Title
University of Guelph

Live body weight (LBW) is an important parameter for both animal health and market requirements for livestock. My thesis focused on exploring the feasibility of a semi-automatic system that estimated the LBW of pigs by applying 4 machine learning (ML) methods (Linear Regression, K-Nearest Neighbor, Decision Tree, and Random Forest) using 3 biometric and 6 morphometric measurements extracted from digital images acquired from 26 pigs in Arkell Swine Research Facility with a consumer-level camera and a reference object. Age, age group, and 6 morphometric measurements significantly influenced LBW estimation, while gender did not. The correlation between manual and image-based measurements was positive and moderately high. Prediction errors for ML models trained on the two acquisition methods were within a  3 kg range, with image-based measurements performing up to 16% lower than manual measurements. This cost- and time-efficient system shows potential for intelligent solutions in animal research and production.

Pig, Machine learning algorithms, Machine learning algorithms, estimation, Live body weight, Machine learning models, semi-automatic/automatic estimation