AUTOMATIC SHEEP BEHAVIOUR ANALYSIS USING MASK R-CNN

The issue of sheep welfare during live exports has triggered a lot of public concern recently. Extensive research is being carried out to monitor and improve animal welfare. Stocking density can be a critical factor affecting sheep welfare during export and its impact can be monitored through sheep behaviour, position, group dynamics and physiology. In this paper we demonstrate the application of the instance segmentation method Mask R-CNN to support sheep behaviour recognition. As an initial step, two typical behaviours standing and lying are recognized under different group sizes in pens over time. 94%+ mAP was achieved in the validation set demonstrating the effectiveness of the method on identifying sheep behaviours. Further data analysis will provide available space requirements for additional sheep allocation and daily behaviour monitoring to detect abnormal cases which will aim to improve the health and wellbeing of sheep on ships.