Files
Abstract
Advancements in sensor technologies and machine learning have revolutionized dairy farming by enabling continuous, automated monitoring of cow behavior, health, and productivity. This thesis integrates multiple data sources, including wearable sensors, automated milking systems, and computer vision, to enhance precision livestock monitoring. A data-driven approach utilizing time-series models was developed to predict milk yield using behavioral, physiological, and environmental variables, improving the accuracy of milk production forecasting. Additionally, a multi-task contrastive learning framework was implemented to enhance individual cow recognition across different behavior classes using small image sets, addressing challenges in computer vision-based livestock monitoring. The proposed methods demonstrated high accuracy in behavior classification, individual identification, and milk yield prediction, offering a scalable and efficient alternative to traditional monitoring approaches. By reducing reliance on labor-intensive observations and improving decision-making efficiency, these technologies provide a powerful tool for optimizing herd management, improving animal welfare, and increasing overall dairy farm productivity.