13TSB_AgriFood: Precision Cow Health Management

Lead Research Organisation: University of the West of England
Department Name: Bristol Robotics Laboratory

Abstract

This project seeks to develop innovative moving 3D imaging technology to enhance measurement of cow body condition score (BCS), live weight and mobility (gait) as a highly advanced management decision-making tool to improve the pace at which these key quality and production traits are identified for animal welfare, sustainability and profitability. The technology will simultaneously and unobtrusively provide stress-free monitoring of incremental changes in individual cow condition and lameness to inform and improve nutrition management, cow health, welfare and productivity leading to increased herd lifespan and more efficient milk production. These traits are currently measured by manual visual assessment requiring high skill levels and training, and so are open to the subjectivity of individuals. Our novel imaging technology will enable much more precise, reliable and frequent measurement, creating greater opportunities to improve cow performance and welfare. Individual cow data will be stored within a central web database to enable benchmarking and on-line dissemination of analysed data back to the farmers and others in the value chain.

Technical Summary

This project seeks to develop innovative moving 3D imaging technology to enhance measurement of cow body condition score (BCS), live weight and mobility (gait) as a highly advanced management decision-making tool to improve the pace at which these key quality and production traits are identified for animal welfare, sustainability and profitability. The technology will simultaneously and unobtrusively provide stress-free monitoring of incremental changes in individual cow condition and
lameness to inform and improve nutrition management, cow health, welfare and productivity leading to increased herd lifespan and more efficient milk production. These traits are currently measured by manual visual assessment requiring high skill levels and training, and so are open to the subjectivity of individuals. Our novel imaging technology will enable much more precise, reliable and frequent measurement, creating greater opportunities to improve cow performance and welfare. Individual cow data will be stored within a central web database to enable benchmarking and on-line dissemination of analysed data back to the farmers and others in the value chain.

Planned Impact

The main beneficiaries are dairy farmers who will be provided with a tool for objectively monitoring cows. Automated Body Condition Score (BCS) and gait measurement, beyond current manual visual assessment, will inform earlier nutrition management (including for breeding) and treatment of lameness respectively. By objectifying and automatically quantifying BCS, the system will enable automatic closed-loop feeding to optimise cow fat presence/condition. This will allow the farmer to easily keep the cow weight/fat and condition at the optimal levels for milk production, as well as avoiding health issues that can be associated with incorrect feeding level/rates. Additionally, the early detection of lameness will allow corrective action to be applied before the condition develops to the extent that milk production is affected or costly vet treatment is needed or the cow needs to be culled. Farmers will benefit from detailed data analysis, benchmarking and reporting through their desired information system as a service. Individual animal data will be integrated with other systems to provide higher level statistics e.g. proxy traits of feed conversion rates. Within a year of the project end, the database of information will be utilised to assess correlations with herd health and genetics, which will help to inform breeding decisions.

A rapid pay back should be achieved by farmers who invest in the system. For a 300 cow herd, yielding 8,000 litres per cow, a typical annual saving through enhanced health management would be £48,500. With a system price of £10,000 and an annual service fee of £5 per cow the return on investment would be 960%. A 5% UK market uptake is expected within 5 years of the end of the project, equating to 725 sales of the system with a gross sales value (at £10,000 per system) of £7.25m together with on-going annual fees from herd benchmarking services amounting to £1.1m (£5/cow/year for 218,000 cows). An annual saving to the industry in excess of £15m would result from savings in net herd replacement cost, improved fertility and reduced lameness. Feedback from producers and industry contacts indicates a strong demand for this type of technology. As well as benefiting dairy farmers by assisting with optimising milk production and cow condition/health, this technology will also provide benefits impacting on other sectors of the dairy industry e.g. vets, nutritionists, milk buyers and retailers. These will arise from greater integration of health management, more productive and profitable dairy herds, improved cow health public image, and enhanced food security (benefits to start emerging within 2 years of the project end). The system also has the potential to be adapted for use by abattoirs for grading both dairy and beef animals prior to slaughter, thereby benefitting that industry.

Manufacture of the systems will provide numerous indirect benefits, such as increased business for system parts suppliers, and in the case of international sales, contributing to UK exports. Internationally, established dairy production is often from larger herds and presents further market opportunities. The top five milk producing countries in the EU have 14 million cows; seven times the UK dairy cow population. Danish and German producers are well known for their early adoption of technology such as robotic milking systems and will be attracted by the benefits of the proposed technology. Even with modest market penetration, sales into the EU of over 1,000 systems could be achieved in 5 years. Rapidly emerging international dairy markets, such as China, India and Russia provide a large potential market into fast developing large scale production units. No such system currently exists in any of the world's dairy producing countries but sales of hi-tech solutions into these markets indicate a willingness to adopt new technology. We believe such international markets should also yield revenues of £1m per annum within 5 years.
 
Description The original proof of concept project established the feasibility of a remote non-contact beta system for use on dairy farms for the monitoring and management of cattle in terms of:
- Animal weight (accuracy within 20kg)
- Animal body condition score (BCS) (more accurately and consistently than a trained observer)
- Animal lameness (correctly identifies 86.6% of lame cows (those with a lameness score of 2 or more)

Key findings include the use of a novel 'rolling ball algorithm' for quantifying animal body condition (essentially how lean/fat the animal is), which the CMV team previously developed for characterising complex 3D morphologies (eg in petrographic analysis of aggregate particles). The CMV team also worked on the use of 3D imaging by, for the first time, tracking and analysing gross body movements as a proxy for conventional leg gait motion in farm animal lameness detection (offering a more practical on-farm solution) and machine learning for animal biometric face recognition. By utilising state-of-the-art deep learning techniques an automated data driven approach allowed a move away from the need for established but unreliable subjective and labour-intensive manual scoring of animals. This body of work led to follow-on BBSRC funding project linked with a KTP to develop a market-ready dairy cow monitoring system, a BBSRC grant, 'Emotional Pig', to further explore applications in farm animal condition monitoring and a linked NERC project to explore the use of dung analysis. The take-home messages for the latter are (1) that it is possible to use deep learning for detection of undigested corn and for large fibres in faeces, and (2) an image processing algorithm (non-machine learning) can score faecal samples according to consistency.

The paper is currently under review for Scientific Reports.

Secondary output is the dataset here: http://researchdata.uwe.ac.uk/481/
Exploitation Route Three follow-on grants have severed to move the product towards commercial application. Agsenze Ltd have attracted further investment (over £300k) and now have the system installed on 11 farms across the UK.
Sectors Agriculture, Food and Drink,Digital/Communication/Information Technologies (including Software)

URL https://www.arlafoods.co.uk/overview/news--press/2019/pressrelease/arla-uk-360-farmers-to-trial-new-3d-imagery-systems-with-automated-intelligence-2895419/
 
Description 'HerdVision' - animal condition monitoring system: improving animal welfare A new company, Agsenze, has been formed (7 jobs created). The 'HerdVision' system was initially developed under this 3-year BBSRC feasibility study grant and then on to a commercial product via follow-on BBSRC and KTP funding. The system was proven at the Agri-EPI South West Dairy Development Centre, a state-of-the-art platform to test and demonstrate new and emerging technologies as both a showcase and testbed for future development (attracting significant reporting across the farming/consumer/brand press including filming for BBC television). A further 11 systems have since been installed on UK farms nationwide supplying Aldi, ASDA and Morrisons - funded by the Origin Group. In 2019 Agsenze Ltd secured an initial £300k of investment from business angels, the Startup Funding Club and Insight Investment, together with KTP funding to develop a new small-format version of the system, now operating in the field - attracting further interest (first reported on by BBC 'Framing Today' and then BBC Six and Ten o'clock news in 2019). This technology offers considerable benefits for framers and wider society by avoiding the need for trained labour intensive and subjective manual observation; instead deploying automation to allow continual twice daily data capture on individual cattle body condition, weight and lameness, linked with milk yield and feed - so offering a mechanism for greater efficiency, increased production / security of production and improved animal welfare. Herdvision is now in use at Arla UK farms as part of the Arla 360 programme, which uses innovative technology to make dairy farming more sustainable. As the UK's biggest producer of dairy, Arla is very concerned to present its products as sustainably and ethically produced. Felicity Callaghan, Head of Media Relations at Arla UK said 'the introduction of Herdvision at Arla 360 farms helps Arla to position itself as an ethical brand'. Sophie Throup, Senior Agricultural Manager at Morrisons, which is supporting the Arla UK 360 farmers programme, welcomed the introduction of Herdvision, commenting 'our customers care about animal welfare, so to know that these trials can improve the well-being of the animals supplying their milk is reassuring'. There are also discussions taking place for overseas trials with LIC, the New Zealand equivalent of Arla in the UK, and in Israel. This project also forms a key component of our REF Case Study.
First Year Of Impact 2019
Sector Agriculture, Food and Drink
Impact Types Economic

 
Description Automated welfare monitoring of dairy cows using 3-dimensional imaging and deep learning, Animal Welfare Seeding Award granted to UWE by Newcastle University on behalf of the BBSRC and the Animal Welfare Research Network
Amount £35,192 (GBP)
Funding ID Animal Welfare Seeding Award granted to UWE by Newcastle University on behalf of the BBSRC and the Animal Welfare Research Network 
Organisation Biotechnology and Biological Sciences Research Council (BBSRC) 
Sector Public
Country United Kingdom
Start 11/2019 
End 02/2020
 
Description FARM interventions to Control Antimicrobial ResistancE (FARM-CARE)
Amount £400,453 (GBP)
Funding ID MR/W031264/1 
Organisation Medical Research Council (MRC) 
Sector Public
Country United Kingdom
Start 02/2022 
End 02/2025
 
Description Grants 4 Growth
Amount £10,000 (GBP)
Organisation University of the West of England 
Sector Academic/University
Country United Kingdom
Start 08/2018 
End 02/2019
 
Description Hoofcount Vision Detection for Early signs of DD Lesions and Lameness Within Dairy Cattle
Amount £418,092 (GBP)
Funding ID 10027119 
Organisation Innovate UK 
Sector Public
Country United Kingdom
Start 04/2022 
End 04/2024
 
Description KNOWLEDGE TRANSFER PARTNERSHIP NO. KTP011688
Amount £182,834 (GBP)
Funding ID KTP011688 
Organisation Knowledge Transfer Partnerships 
Sector Charity/Non Profit
Country United Kingdom
Start 03/2019 
End 03/2021
 
Description Knowledge Transfer Partnership
Amount £211,075 (GBP)
Funding ID KTP010689 
Organisation Innovate UK 
Sector Public
Country United Kingdom
Start 05/2017 
End 05/2019
 
Description Knowledge Transfer Partnership (KTP)
Amount £171,094 (GBP)
Organisation Innovate UK 
Sector Public
Country United Kingdom
Start 05/2017 
End 06/2019
 
Description Pig ID: developing a deep learning machine vision system to track pigs using individual biometrics
Amount £612,000 (GBP)
Funding ID BB/X001385/1 
Organisation Biotechnology and Biological Sciences Research Council (BBSRC) 
Sector Public
Country United Kingdom
Start 02/2023 
End 01/2025
 
Description Sustainable Agriculture Innovation Call May 2016
Amount £195,386 (GBP)
Funding ID NE/P007996/1 
Organisation Natural Environment Research Council 
Sector Public
Country United Kingdom
Start 01/2017 
End 07/2018
 
Description Sustainable Agriculture Innovation Call May 2016
Amount £199,999 (GBP)
Funding ID NE/P007945/1 
Organisation Natural Environment Research Council 
Sector Public
Country United Kingdom
Start 09/2016 
End 09/2018
 
Description Using computer vision to determine emotion in animals
Amount £90,000 (GBP)
Organisation University of the West of England 
Sector Academic/University
Country United Kingdom
Start 03/2022 
End 04/2025
 
Description the GCRF Agri-tech Catalyst Seeding Award competition.
Amount £142,000 (GBP)
Funding ID GCRF-SA-2020-UWE 
Organisation Biotechnology and Biological Sciences Research Council (BBSRC) 
Sector Public
Country United Kingdom
Start 03/2020 
End 07/2021
 
Title Technologies for assessment of nutrient digestibility in cattle 
Description Accurate feeding of animals in the beef and dairy industries is important both for efficient production and to reduce the impact of cattle farming on the wider environment. Both under and over feeding of nutrients are inefficient and can lead to environmental, economic and welfare issues. Farm businesses cannot afford to waste expensive resources by feeding nutrients in amounts surplus to requirements. Equally it is not uncommon for farm rations to perform under expectations; too much or too little of some components, poor mixing, or sorting can lead to poor productivity and health and welfare problems. Key issues are presence of excess starch or too little effective fibre in the feed. However, the farmer's ability to make feed strategy decisions quickly is restricted by the time needed for off-site lab-based chemical analysis of feed and the lack of an appropriate method for determining the appropriate level of dietary effective fibre. Near infra-red reflectance spectroscopy (NIRS) is a technique used by commercial analytical laboratories to estimate forage and feed quality from prior cross-reference with calibration algorithms derived from standard chemical analyses. This project aims to transfer the analysis from a time consuming, laboratory based approach to a real time, farm-based diagnostic. Improving the accuracy of predictions provided by hand held NIR spectral analysis of feed and faeces is critical to the development of a system for rapid and accurate assessment of feed utilisation that can be applied on-farm. We aim to overcome major technical challenges associated with making accurate predictions of feed quality and digestibility by combining expertise in animal nutrition and computational image analysis. The first challenge was to understand limitations to reproducibility in NIR spectra. NIR is most typically used for dry samples and water in the sample introduces variation and errors. A measurement system has been devised in which a wet sample can be accurately detected by the AUNIR NIR4 Farm probe, without damaging the probe (designed for dry samples) and allowing measurement of a representative sample size. We subsequently applied this methodology to investigate the effect of time and multiple freeze-thaw cycles on NIR spectra generated from faecal samples collected during digestibility trials. Repeated NIR measurements were made on samples from a cattle feeding experiment (8 cows, 2 on each of 4 diets). NiR4farm was used to measure fresh samples, which were then subdivided and frozen. At intervals between 1 and 10 weeks, samples were thawed, measured and re-frozen such that replicate samples were measured once or at multiple times. Multivariate analysis of NIR spectra was conducted and revealed no effect of time or number of freeze/thaw cycles on the data. However, the underlying experimental treatment structure of the original feeding trial experiment was detected. This justifies the use of frozen samples for the next stage of development. Here we are analysing samples from current experimentation for more detailed analysis of the relationships between NIR spectra and measurements of key traits relating to feed quality (nitrogen, starch, water soluble carbohydrate, fibre, digestibility and diet particle size), and animal performance (dry matter intake, nutrient digestion, live weight gain and milk yield). Our second approach, to which this dataset pertains, involves image analysis as a less subjective version of the on farm "boot test" of faecal consistency as an indicator of gut health and 'effective fibre' content of the diet. A portable imaging system has been developed to include NIR or visible light sources. By using faecal simulations, 3D imaging techniques (photometric stereo) have been evaluated to identify key features in cattle faeces indicative of digestive health (e.g. consistency, presence of fibres). Machine learning has been applied to training images to extract data related to presence of fibres and corn kernels in the samples. Further algorithms were developed to describe the "roughness" of the surface of fresh faecal samples. It is envisaged that ultimately these tools will allow derivation of accurate calibration algorithms from which it will be possible to estimate quality of feed inputs and extent of utilisation (from the corresponding faecal measures). In combination with improved NIRS, the development of 'on-farm visual analysis' as an additional automated diagnostic tool will significantly enhance the ability to make real-time feeding decisions. This will enable more precise, strategic feeding of individual animals and herds for on-farm feed and manure nutrient management to improve welfare, production and the environment. 
Type Of Material Database/Collection of data 
Year Produced 2019 
Provided To Others? Yes  
Impact Take-home messages are (1) that it is possible to use deep learning for detection of undigested corn and for large fibres in faeces, and (2) an image processing algorithm (non-machine learning) can score faecal samples according to consistency. A paper is currently under review for Scientific Reports. 
URL http://researchdata.uwe.ac.uk/481/
 
Description Agsenze Ltd 
Organisation Agsenze Ltd
Country United Kingdom 
Sector Private 
PI Contribution Technical expertise
Collaborator Contribution Commercial expertise
Impact New funding attracted: Grants4Growth; KTP
Start Year 2018
 
Description Kingshay Farming & Conservation Ltd 
Organisation Kingshay Farming & Conservation Ltd
Country United Kingdom 
Sector Private 
PI Contribution Technical expertise
Collaborator Contribution Application expertise
Impact Patent Papers
Start Year 2013
 
Title Patent 
Description Patent 
IP Reference GB1505461.2 
Protection Patent application published
Year Protection Granted 2015
Licensed Yes
Impact Technology nearing a commercial stage for automated cattle condition monitoring.
 
Title HerdVision 
Description New technology and software realised for estimating the body condition and mobility scoring of dairy cows on the farm. Real-time data is presented to the farmer continuously. It is sold under the brand name of 'HerdVision'. 
Type Of Technology New/Improved Technique/Technology 
Year Produced 2021 
Impact Farmers using the system report significant benefits relating to animal welfare and farm management (see testimonials available via the URL). 
URL https://herd.vision/
 
Description "Computer vision and deep learning for on-farm welfare assessments of dairy cows and pigs", Invited speaker at The 7th International Conference on Animal Computer Interaction 2020. November 2020 
Form Of Engagement Activity Participation in an activity, workshop or similar
Part Of Official Scheme? No
Geographic Reach National
Primary Audience Professional Practitioners
Results and Impact Mark Hansen gave an invited talk on our work in "Computer vision and deep learning for on-farm welfare assessments of dairy cows and pigs", at The 7th International Conference on Animal Computer Interaction 2020.
Year(s) Of Engagement Activity 2020
URL http://www.aciconf.org/aci2020/program
 
Description BBC Radio 4 Farming Today 
Form Of Engagement Activity A press release, press conference or response to a media enquiry/interview
Part Of Official Scheme? No
Geographic Reach National
Primary Audience Public/other audiences
Results and Impact An item on our IUK funded 'How's my Cow?' appeared on BBC Radio 4's Farming Today program that include an interview with a member of the project team.
Year(s) Of Engagement Activity 2015
URL http://www.bbc.co.uk/programmes/b065t315
 
Description Gave an interview to New Scientist for an article entitled "Smart dairy farms are using AI scanners to monitor cows' health". 
Form Of Engagement Activity A press release, press conference or response to a media enquiry/interview
Part Of Official Scheme? No
Geographic Reach International
Primary Audience Media (as a channel to the public)
Results and Impact Gave an interview to a science journalist specialising in animal health and behaviour, Christa Lesté-Lasserre, from which an article appeared on 31 January 2023 in New Scientist, entitled "Smart dairy farms are using AI scanners to monitor cows' health".
Year(s) Of Engagement Activity 2023
URL https://www.newscientist.com/article/2357313-smart-dairy-farms-are-using-ai-scanners-to-monitor-cows...
 
Description Invited speaker (Mark Hansen) at the Animal Welfare Research Network workshop in Thermography in Edinburgh this September. 
Form Of Engagement Activity A talk or presentation
Part Of Official Scheme? No
Geographic Reach National
Primary Audience Industry/Business
Results and Impact Mark was an invited speaker at the Animal Welfare Research Network workshop in Thermography in Edinburgh this September. The talk was very well received and generated a large amount of interest in machine vision.
Year(s) Of Engagement Activity 2018
 
Description Item on region news 
Form Of Engagement Activity A press release, press conference or response to a media enquiry/interview
Part Of Official Scheme? No
Geographic Reach Regional
Primary Audience Media (as a channel to the public)
Results and Impact Item on HerdVision, including mention of a technology award, appeared on BBC Points West regional news.
Year(s) Of Engagement Activity 2022
URL https://www.facebook.com/watch/?v=995406818522443
 
Description Items appeared on our work with pigs and dairy cattle as part of the' future of farming' on BBC Six and Ten o'clock news, digital (online video) and Newsround. 
Form Of Engagement Activity A press release, press conference or response to a media enquiry/interview
Part Of Official Scheme? No
Geographic Reach National
Primary Audience Public/other audiences
Results and Impact Television interviews recorded on both our work in pig face and emotion recognition, jointly with SRUC, and also our on-going work in condition monitoring of dairy cattle.
Follow-up interest occurred both from media organisations and others (academic and commercial) interested in potential collaboration.
Other media interest included:

Item recorded for Netflix series "Connected" in September 2019 to be broadcast summer 2020.

Daily Mail: https://www.dailymail.co.uk/wires/pa/article-6821859/Facial-recognition-technology-used-discover-emotional-state-pigs.html

New Food Magazine: https://www.newfoodmagazine.com/article/92909/researchers-develop-machine-vision-technology-to-detect-a-pigs-emotional-state/

Imaging and Machine Vision Europe: https://www.imveurope.com/news/pig-expressions-studied-face-recognition

SPIE - Optics - https://optics.org/news/10/3/35
Year(s) Of Engagement Activity 2019
URL https://www.bbc.co.uk/news/av/science-environment-49362428/pigs-emotions-could-be-read-by-new-farmin...
 
Description KTN Emerging Imaging Technologies in Agri-Food Workshop in Birmingham 
Form Of Engagement Activity Participation in an activity, workshop or similar
Part Of Official Scheme? No
Geographic Reach National
Primary Audience Industry/Business
Results and Impact Wenhao Zhang presented our work on using machine vision / learning in agri-tech at the KTN Emerging Imaging Technologies in Agri-Food Workshop in Birmingham
Year(s) Of Engagement Activity 2018
URL https://ktn-uk.co.uk/news/emerging-imaging-technologies-in-agri-food
 
Description KTN Networking evenet - Robotics and Artificial Intelligence for Agriculture Event, 26th July (London) 
Form Of Engagement Activity A talk or presentation
Part Of Official Scheme? No
Geographic Reach National
Primary Audience Industry/Business
Results and Impact showcasing some Recent successful InnovateUK projects for AgriFood.

Lots of interest in our work.
Year(s) Of Engagement Activity 2018
URL https://ktn-uk.co.uk/events/19061
 
Description Virtual presentation to UK-RAS Strategic Task group in Agri-Robotics 
Form Of Engagement Activity A talk or presentation
Part Of Official Scheme? No
Geographic Reach National
Primary Audience Professional Practitioners
Results and Impact Melvyn Smith to give a virtual presentation to UK-RAS Strategic Task group in Agri-Robotics that included our work on automated weed detection, cattle condition monitoring and pig face recognition and expression detection on 29th Sept.
Year(s) Of Engagement Activity 2020
 
Description White paper - Agricultural Robotics: The Future of Robotic Agriculture 
Form Of Engagement Activity A formal working group, expert panel or dialogue
Part Of Official Scheme? No
Geographic Reach National
Primary Audience Policymakers/politicians
Results and Impact UK-RAS Network White Papers, ISSN 2398-4414
Year(s) Of Engagement Activity 2018
URL https://arxiv.org/abs/1806.06762