13TSB_AgriFood: Automated screening for pathologies at abattoir through computer vision- based inspection of pig carcasses

Lead Research Organisation: University of Dundee
Department Name: Computing

Abstract

Visual inspection of carcasses is an important factor for ensuring the quality of meat products. However, manual inspection puts a strain on meat inspector resources, which effectively prevents detailed screening as, for example, for the purposes of health schemes.
The aim of this project is to develop an automated system for visual screening at abattoir. The system will analyse images taken at abattoir in order to detect a number of health hazards on pig carcasses, and to screen them at slaughter for indications of underlying subclinical diseases. The project will: (1) develop a system for capturing images of carcass; (2) acquire image data sets and have experts annotate pathologies in them; (3) develop software that learns to recognise pathologies automatically; (4) validate the system on large scale datasets for the detection of routine health hazards, and disseminate results to relevant users, incl. providing feedback to farmers.
The project brings together market leaders in meat production (Tulip) and supplier for abattoirs (Hellenic), pig levy board (BPEX), the UK's leading centre for research into pig science (Newcastle Univ.), and experts in computer vision and pattern recognition (Newcastle Univ. and Univ. of Dundee). It will enhance confidence in detecting health hazards in pig carcasses, aiming towards automated detection of underlying subclinical disease. Feeding this back to pig farmers will increase productivity and improve efficiency on farms through preventing further diseases. Producers will be able to make decisions about improving the health of their herd through the information they receive from the abattoir. The project will thus contribute towards sustainability and competitiveness of the UK pig Industry.

Technical Summary

The aim of the proposed project is to develop an automated system, based on visual image analysis that detects a number of public health hazards on pig carcasses through carcass inspection at abattoir, and to screen pig carcasses at slaughter for conditions indicative of subclinical diseases. The technical objectives of the proposal are to: (1) develop and deploy multi-camera recording infrastructure that enables capture of images of carcasses; (2) acquire image data sets and have experts annotate pathologies in these images; (3) develop and refine algorithms/software that can automatically recognise pathologies; (4) validate these algorithms on large scale datasets, and disseminate results to relevant users. These aims align with the TSB competition aim of 'Data capture and integration through the value chain'. The automated capture of data associated with subclinical diseases will be fed to pig producers through the activities of BPEX Pig Health Scheme; producers will then be able to make decisions about improving the health of their herd. As such, the proposal is also directly aligned to the competition aim of 'Management decision making'.

Planned Impact

Impact on pork industry
Meat inspection is a significant cost to the industry, and regardless of the size of the abattoir these costs are still incurred.
The introduction of the proposed automated system into the abattoir inspection systems will reduce such costs.
Outside the participating Industry, the primary beneficiaries of this research will be the wider food production industry that
does not participate in the project: the broader Pig Sector, the Government and Government Agencies, the Academic
community and the Wider Public who will benefit as a consequence of successful implementation.
Industry partners in the consortium will be in the unique position of exploiting the outcomes of the project for the benefit of
their businesses. This will confer major competitive advantages, especially as outcomes will be applicable to the wider pig
sector, and potentially beyond. In the first instance Tulip's abattoir in Spalding, which will serve as development facility, will
benefit from the automated analysis system. Subsequently the system can be rolled out to all Tulip facilities, which will
result in automatic screening of effectively ~30% of the UK's pig kill. With moderate site-specific modifications the proposed
technology can be deployed in most abattoirs (world-wide). The system could be potentially installed in all pig abattoirs in
the UK and the European Union, provided that it obtains approval from the (E)FSA.
Wider governmental and societal impact
BPEX involvement in the project will ensure potential adoption of the product by all abattoirs that slaughter pigs in the UK.
The current mechanisms of BPHS for disseminating information on the incidence of pathologies to pig producers will
ensure that the generated information will reach the majority of UK pig producers.
In general the outcomes of the project will be of particular relevance to policy makers, especially to those that aim to ensure
production of safe and high quality meat products. In general, there is increased public interest in the safety of livestock
products, including authentication and incidence of zoonotic diseases. Processors and retailers are the conduit for such
concerns. Part of the project outcomes will be the production of safer pork products, which is of interest to retailers and
public alike. FSA publishes reports about the incidence of such pathologies on its websites and retailers are very keen to
disseminate information about the safety of their products.
Academic Impact
Academic partners will benefit from the application of their skills to pig production systems, and the challenges this brings,
and the consequent enhancement of their on-going research. Potentially the algorithms developed for the detection of
pathologies in pig carcasses could be extended to detect pathologies in other livestock carcasses. This will significantly
expand the scope of the research of the academic partners. The application of the developed algorithms to pig pathologies
will be subject to communication at the latter stages of the project.
Exploitation and Application
Project outcomes are expected to confer competitive advantages on the industry partners involved. Information generated
on the incidence of specific pathologies will directly benefit Tulip-associated pig producers and eventually all BPHS
associated pig producers. The application of the automated system will increase the number of carcasses inspected and
thus enhance the value of the BPHS. Hellenic Systems would be expected to market the automated recording systems to
UK abattoirs, including those associated with Tulip.
Clearly the impact of the product developed by the project would be enormous if the system is adopted by (European) FSA.
The impact activities will be primarily undertaken by BPEX, Tulip and the academic Institutions; all of them have
considerable relevant KT experience and successes as outlined above.

Publications

10 25 50
 
Description The technical objectives of the proposal were: (1) to develop and deploy multi-camera recording infrastructure that enables capture of images of carcasses, suitable for automated analysis; (2) to acquire image data sets and have experts annotate pathologies in these images; (3) to develop and refine algorithms/software that can automatically recognise pathologies; and (4) to validate these algorithms on large scale datasets, and disseminate results to relevant users.
All technical objectives of the proposal were met, although the consortium decided to focus on detection of pathologies on offal rather than on external carcass pathologies, as the majority of the pathologies considered to pose public health issues are associated with the former.

We have developed a method for automated detection of localised pathologies in pig offal at abattoir. These pathologies are either spot like, i.e. localised lesions or diffuse, i.e. extended across an offal. They are associated with the major conditions detected on pigs at UK abattoirs. The key findings of the project were: 1. We developed hardware that enabled us to collect pig carcass and offal pathologies pictures at abattoir. This overcame the usual challenges that arise from a hostile environmental, such as an abattoir, and of obtaining images from object that are three dimensional and may be occluded. 2. We captured images of pathologies on carcass and offal and create a training library (annotated images). 3. We developed machine learning algorithms for automated screening of pathologies; we focused on two types of pathologies, milk spots and pericarditis. These two pathologies represent examples of spot like and diffuse pathologies respectively; at the same time they represent pathologies that have high prevalence amongst pig carcasses. 4. We applied the machine learning algorithms on the automated detection of pneumonia, which can be considered as a diffuse pathology respectively; the accuracy of detection for this pathology was also > 80%. 5. We validated the developed methods on independent data; the current accuracy of the detected pathologies was between 82-89% (the lower being for the milk spots, the high for pericarditis). 6. The outcomes of the project can be used as the basis of an automated detection method of pathologies of pigs at abattoirs in practice.
Exploitation Route This was an Innovate UK project and the findings are of Industrial Relevance; the algorithms are proprietary. The achievements of the project will contribute enormously towards the development of automated detection of pig pathologies at abattoir. We have clear plans about how to develop further the technology in order to enhance the usability of the method. This will require the acquisition of additional images of pathologies of low prevalence. We have plans for collaboration with providers of such images, but this will imply that we will need to obtain the images from pigs with a lower health status.
Sectors Agriculture, Food and Drink

 
Description We have developed algorithms that enable the detection of two types of pathologies in pig carcasses; both pathologies are associated with viscera. These algorithms are expected to lead to the development of an automated method for the screening of pig pathologies at abattoir; this development is the responsibility of the Industry partners associated with this BBSRC and IUK award. The algorithms can be extended to the automated detection of other pathologies, which are either spot -like or diffuse. In order to do this we would need to acquire significant images to train the algorithms to achieve this.
First Year Of Impact 2018
Sector Agriculture, Food and Drink
Impact Types Economic

 
Description Collaborative project with industry (Innovate UK) and Newcastle University (BBSRC) 
Organisation Hellenic Systems Ltd
Country United Kingdom 
Sector Private 
PI Contribution I am part of a consoritum funded by Innovate UK and BBSRC which involves Tulip Ltd., Hellenic Systems Ltd., AHDB, and Newcastle University. My input is largely on development of computer vision and machine learning methods for the project.
Collaborator Contribution Industry input such as data acquisition. Veterinary input including data annotation. Software development.
Impact See outputs from 13TSB_AgriFood: Automated screening for pathologies at abattoir through computer vision- based inspection of pig carcasses
Start Year 2014
 
Description Collaborative project with industry (Innovate UK) and Newcastle University (BBSRC) 
Organisation Newcastle University
Country United Kingdom 
Sector Academic/University 
PI Contribution I am part of a consoritum funded by Innovate UK and BBSRC which involves Tulip Ltd., Hellenic Systems Ltd., AHDB, and Newcastle University. My input is largely on development of computer vision and machine learning methods for the project.
Collaborator Contribution Industry input such as data acquisition. Veterinary input including data annotation. Software development.
Impact See outputs from 13TSB_AgriFood: Automated screening for pathologies at abattoir through computer vision- based inspection of pig carcasses
Start Year 2014
 
Description Collaborative project with industry (Innovate UK) and Newcastle University (BBSRC) 
Organisation Tulip Ltd
Country United Kingdom 
Sector Private 
PI Contribution I am part of a consoritum funded by Innovate UK and BBSRC which involves Tulip Ltd., Hellenic Systems Ltd., AHDB, and Newcastle University. My input is largely on development of computer vision and machine learning methods for the project.
Collaborator Contribution Industry input such as data acquisition. Veterinary input including data annotation. Software development.
Impact See outputs from 13TSB_AgriFood: Automated screening for pathologies at abattoir through computer vision- based inspection of pig carcasses
Start Year 2014
 
Description BPHS Steering Group meeting (Kenilworth) 
Form Of Engagement Activity A talk or presentation
Part Of Official Scheme? No
Geographic Reach National
Primary Audience Industry/Business
Results and Impact This was a Steering Group meeting of the British Pig Health Scheme (BPHS) at the Agriculture and Horticulture Development Board (AHDB) office in Stoneleigh Park.
Year(s) Of Engagement Activity 2016
 
Description Invited Presentation at the Pig Veterinary Society Annual Meeting - Leeds 
Form Of Engagement Activity A talk or presentation
Part Of Official Scheme? No
Geographic Reach National
Primary Audience Professional Practitioners
Results and Impact An invited presentation of the project outcomes was made at the Pig Veterinary Society Conference in November 2017. The Conference was attended by all veterinarians who deal with pig health issues, both academic and practitioners. The conference was attended by approximately 110 delegates. The presentation was timely given the requirements of the European Food Standards Agency for visual only inspection of pig carcasses. There has been increased interest to pursue further the automated detection of pig pathologies at abattoir. Possibilities for setting up collaborations to address this are currently in progress.
Year(s) Of Engagement Activity 2017
URL https://www.pigvetsoc.org.uk/files/document/894/1711%20PVS%20Programme.pdf
 
Description Presentation at workshop for industry stakeholders 
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 A workshop held at University of Lincoln was to present findings to industry stakeholders on potential technologies for automated meat
inspection, targeted specifically at issues relating to public health on red meat and poultry.
Year(s) Of Engagement Activity 2017
 
Description Stakeholder Conference 
Form Of Engagement Activity A talk or presentation
Part Of Official Scheme? No
Geographic Reach International
Primary Audience Industry/Business
Results and Impact A presentation was made to an Industry event organised by a major pharmaceutical company for its veterinarians and clients. The presentation aimed at introducing the idea of automated detection of pig carcasses at abattoir to a wider audience, which eventually might be the end users. The presentation was the focus of the conference which aimed at introducing precision farming in the pig sector.
Year(s) Of Engagement Activity 2018
 
Description Talk and abstract at IPVS 2016 
Form Of Engagement Activity A talk or presentation
Part Of Official Scheme? No
Geographic Reach International
Primary Audience Industry/Business
Results and Impact Talk and abstract at 24th International Pig Veterinary Society Congress title "Image analysis for automated detection of abnormal organs in pig offal" by Telmo Amaral1, Thomas Plötz1, Stephen McKenna2, Terry Carter3, Katharine Yuill4, Jen Waters4, Ilias Kyriazakis* 5
1Open Lab, School of Computing Science, Newcastle University, Newcastle upon Tyne, 2CVIP, Computing, School of
Science and Engineering, University of Dundee, Dundee, 3Hellenic Systems Ltd, South Woodham Ferrers, 4Tulip Ltd,
Warwick, 5School of Agriculture, Food and Rural Development, Newcastle University, Newcastle upon Tyne, United
Kingdom
Year(s) Of Engagement Activity 2016
URL http://www.ipvs2016.com/
 
Description Workshop on Machine Vision of Animals and their Behaviour 
Form Of Engagement Activity Participation in an activity, workshop or similar
Part Of Official Scheme? No
Geographic Reach International
Primary Audience Industry/Business
Results and Impact A workshop with a programme featuring contributions affiliated to 13 universities, two companies and one governmental body.
Year(s) Of Engagement Activity 2015
URL http://openlab.ncl.ac.uk/publicweb//mvab2015/