Prof Christos Tachtatzis, University of Strathclyde; Prof Gareth Arnott, Queen's University Belfast; Steven Morrison and Francis Lively, Agri-Food and Biosciences Institute (AFBI)

Social interaction and spatial organisation are fundamental components of welfare in group-housed dairy cattle, shaping access to resources, behavioural synchrony, and responses to management and environmental conditions.

 However, social behaviour is difficult to measure at scale in commercial systems, and current approaches rely largely on manual observation or coarse indicators that do not capture group dynamics. Recent advances in computer vision offer the potential to quantify social behaviour continuously and non-invasively, but methods that are robust, interpretable, and suitable for deployment in real farm environments remain limited. 

This PhD project will develop vision-based machine learning methods for quantifying social interaction, spatial organisation, and collective behavioural patterns in housed dairy cattle. Using video data collected from multi-camera systems deployed in commercial-scale housing, the student will develop identity-preserving tracking and interaction detection methods capable of operating over long time periods and in high-density environments. 

These methods will be used to extract welfare-relevant social and behavioural metrics, including proximity networks, affiliative and avoidance behaviours, displacement events, dominance-related interactions, and behavioural synchrony. Machine learning models will integrate these behavioural signals to infer group-level structure and welfare-relevant states, with a strong emphasis on biological interpretability and validation. Validation will be supported through expert behavioural interpretation and comparison with established animal science knowledge, ensuring that outputs are meaningful and relevant to applied livestock research and management. 

The project will be conducted in close collaboration with academic and industry partners and will use data collected under realistic operating conditions. The student will receive interdisciplinary training in computer vision, machine learning, and applied livestock research, alongside experience of industry engagement and knowledge transfer.

The outcomes will support the development of interpretable, vision-based tools for assessing social behaviour and welfare in sustainable dairy production systems.