AI-driven analysis of cattle behaviour in lairage: Impact of flooring substrate and implications for welfare and management

Lairage time for beef cattle is the short period following transportation where the animals are confined in pens before being taken to slaughter. Optimising the environment during this period is critical for meeting animal welfare requirements as well as maintaining optimal meat quality. However, there is high animal throughput, day and night, presenting a challenge for personnel to visually monitor all animals within these facilities. Supportive computer vision tools would improve the ability to continuously monitor animals for timely interventions as needed or for informing housing and husbandry changes. The objectives of this research were to develop computer vision algorithms to automatically classify cattle behaviours in lairage pens at a commercial facility that differed in their flooring substrate. Manually annotated video frames of 24,619 instances of individual cattle behaviours from recordings taken in a commercial cattle lairage facility in Australia were used to develop a deep learning framework. This framework employed a two-stage approach incorporating object detection, tracking, and motion analysis to categorise cattle behaviours into three primary states: lying, standing, and walking (animals in motion). The final model evaluation on 82 validation images containing 2,427 cattle behaviour instances showed a classification accuracy of 97.7% between lying and standing/walking. Real-time analysis and time-period proportion statistics of 18 groups of cattle were used to evaluate behavioural differences between cattle housed on concrete flooring or woodchip bedding. There was a significant interaction between group and flooring substrate type on standing, walking, and lying behaviour (all P ≤ 0.001) with more standing, less walking, and less lying in the animals housed on concrete substrate but this difference varied across the groups. The application of computer vision monitoring in lairage can inform on housing and management changes to improve animal welfare. Future research should aim to understand the factors behind inter-group variation, specific to the lairage context, as a critical next step in for meaningful interpretation of automated behavioural classifications. Developing a robust commercial detection system with behavioural thresholds could aid in optimising animal welfare and meeting regulatory auditing requirements. • Automated behaviour detection in lairage can improve cattle monitoring • Developed computer vision algorithms classified lying, standing, walking • Cattle behaviour in commercial lairage pens was automatically processed • Cattle behaviour differed between woodchip bedding and concrete flooring • There was significant inter-group variation in behaviours