Developing AI-enhanced continuous monitoring of individual locomotor dynamics to detect shifts in poultry behaviour, health and welfare in group housi
Lead Research Organisation:
Royal Veterinary College
Department Name: Pathobiology and Population Sciences
Abstract
There is an urgent need for objective, evidence-based tools to monitor animal health and welfare. Locomotor dynamics offer promising opportunities to quantitatively and continuously track individual animal behaviour, to improve management of health and welfare. This project aims to use computer vision and machine-learning (or artificial intelligence, AI) to detect and classify poultry motion within group-housing. Automation of behaviour monitoring has commercialisation potential in precision livestock farming, treatment response monitoring of research animals (vaccine and drug development), and companion animal healthcare. This project aligns well with Key Challenges in the BBSRC strategic framework, including One Health and Biotechnology for Health.
Organisations
| Description | Monitoring of poultry welfare outcomes allows early detection of issues, aiding prevention and reducing suffering. Accelerometers are increasingly being investigated to detect animal behaviour as a method for monitoring individual welfare that is time-saving, continuous and objective. Overall, this thesis aimed to: i) identify relevant behavioural and accelerometer indicators (accelerometer activity: percentage of time spent active (activityA)) of welfare states (lameness and hock burn) and disease state (coccidiosis), as key poultry welfare states, in different strains of broiler chickens under challenge (disturbed and carrying accelerometers); and ii) to develop random forest models to detect selected behaviours and gait score. To investigate welfare states, three strains of broilers were compared: a fast-growing conventional strain (FG-R) and two slower-growing strains (SG-H; SG-NB). Firstly, FG-R had poorer gait score compared to slower-growing strains and compared to FG-R, slower-growing strains sat inactive and sat preening for shorter durations and walked for longer durations. These behaviours were reflected in activityA which was significantly lower in FG-R compared to slower-growing strains. ActivityA was negatively associated with weight but, counter to predictions, not hock burn or poor gait score. In the diseased birds, counter intuitively, infected birds sat for shorter durations and performed longer average bout durations of wing assisted running. This was mirrored in activityA, which significantly decreased in non-infected birds but did not reflect differences in coccidiosis lesions scores (post-mortem intestinal lesions) measuring infection severity. Lastly, using random forest algorithms and accelerometer attributes, sitting, standing, and walking behaviours were classified with high accuracy. Algorithms also showed potential for classifying specific gait scores across multiple broiler strains. The findings demonstrate that locomotor dynamics, recorded using accelerometers, serve as a continuous and quantitative metric of behaviour and have the potential to be used as automated indicators of both welfare and disease at the individual level. Note: As well as the BBSRC, this work was co-funded by the Royal Society for Prevention of Cruelty to Animals (RSPCA) and Open Philanthropy. |
| Exploitation Route | With further research a tool could be developed to provide a quantitative, less subjective measure of lameness than currently stands. The results of this study also suggest that individual activity measured by accelerometers, whilst accounting for change with age, could act as a warning signal for the development of lameness. With further research, there is the potential to develop an automatic tool to monitor behaviour and welfare issues such as lameness. |
| Sectors | Agriculture, Food and Drink |
| Description | 3MT Competition |
| Form Of Engagement Activity | A talk or presentation |
| Part Of Official Scheme? | No |
| Geographic Reach | Regional |
| Primary Audience | Other audiences |
| Results and Impact | The 3 minute thesis competition enabled me to summarise my thesis with only 3 minutes. It's purpose was to practise public engagement and describing why the work is important to a lay audience. |
| Year(s) Of Engagement Activity | 2020 |
| Description | Poultry Research Day |
| Form Of Engagement Activity | A talk or presentation |
| Part Of Official Scheme? | No |
| Geographic Reach | International |
| Primary Audience | Other audiences |
| Results and Impact | A poultry research day was organised by Bristol University for academics from around the world to introduce and discuss their current or previous research in related fields. It was interesting to hear from people who had faced similar challenges to the ones I was facing in my research at the time. |
| Year(s) Of Engagement Activity | 2019 |
| Description | RVC Postgraduate Research Day |
| Form Of Engagement Activity | Participation in an open day or visit at my research institution |
| Part Of Official Scheme? | No |
| Geographic Reach | Local |
| Primary Audience | Postgraduate students |
| Results and Impact | I presented a powerpoint talk as well as a poster on two separate occasions as part of my 1st year at my research institute. Attendees included postgraduate students and academics/lecturers. It was a good experience to hear and answer questions which are likely to arise with this research. |
| Year(s) Of Engagement Activity | 2019 |