Investigating automatic detection of emotion in biometrically identified pig faces using machine learning

Early identification and resolution of pig health issues results in reduced production costs and improved animal wellbeing. Current manual visual inspection offers only intermittent and subjective information often at a group level. A capacity for continuous automated monitoring of individual pigs allows on-going learning about individuals, and consequently allows early detection of altered health and welfare, so permitting more timely and cost effective remedial interventions. Going beyond that goal would be the development and harnessing of technology capable of assessing animal affective state thus offering a truly insightful, animal-centric welfare assessment tool. This research is particularly novel and timely as it uses highly innovative technologies to develop an animal-centric assessment of welfare, understanding that the sentience of animals is something of great importance to society and policy makers. By focussing on the highly individual measure of facial expression we can deliver a welfare assessment technique that goes beyond basic monitoring to actually inferring something about the importance the animals themselves place on particular experiences. Whilst the absence of negative affective state is and should be a priority we will also include particularly novel work to detect positive affect in facial expression, thus moving closer to the ultimate goal of measuring whether animals are experiencing "a good life". The project is therefore relevant to the BBSRC strategic priorities: welfare of managed animals, sustainably enhancing agricultural production, animal health and technology development for the biosciences. Machine vision offers the potential to realise a low-cost, non-intrusive and practical means to both biometrically identify individual animals and then assess and record their condition continuously each day using only the face. By employing state-of-the-art machine learning techniques, such a system would offer the capacity for on-going learning about individuals, and consequently allow for early detection of altered health/welfare, personalised thresholds for intervention, and tailored treatment approaches. Such individualised data recording can also be used in a wider precision farming context by association with other measurable parameters, such as individual food and water intake, treatment history, growth and weight gain, in order to better optimise farm production efficiency. The most innovative application, however, is the potential to use a non-intrusive technique to infer affective state, allowing insight into both short-term emotional reactions and longer-term individual "moods" of animals under human care. The project delivers clear benefits to multiple stakeholders within the livestock sector and therefore it has attracted the interest and support of influential industry and technology companies keen to be involved, from the start, in the development and application of such an innovative approach to animal welfare assessment.