Weighing finishing pigs in motion: A walk-over scale for accurate weight estimation

Accurate and efficient weight estimation of pigs is crucial for optimizing production, ensuring animal welfare, and making informed decisions in swine farming. Despite technological advancements, obtaining precise individual pig weights remains challenging due to the dynamic nature of pig movement and the stress induced by traditional weighing methods, highlighting the need for innovative, non-invasive solutions. This study presents an automated walk-over scale system that leverages high-frequency load cell data, feature engineering, and machine learning techniques to estimate pig weights in motion, addressing the limitations of traditional weighing methods. The system’s effectiveness was validated in a real-world setting with 50 pigs across 944 walk-throughs, achieving a Root Mean Square Error (RMSE) of 2.87 kg and a Mean Absolute Percentage Error (MAPE) of 2.65% on a 20% pig-wise holdout validation set, demonstrating its potential as a practical solution for non-invasive, accurate weight monitoring in commercial pig farming operations. • Developed an automated walk-over scale for dynamic pig weight estimation. • Used high-frequency load cell data with machine learning for weight predictions. • Improved weight prediction with zero-acceleration weights and segmented features. • Achieved an RMSE of 2.87 kg and a MAPE of 2.65% in real-world farm trials. • Gradient Boosting Regressor was the most effective for predicting the weight of pigs.