Page 10 - National Poultry Newspaper

Summary image of machine vision techniques for performing automated monitoring of flock motion, weight estimation and behaviour classification.
Cheryl McCarthy and Derek Long of the University of Southern Queensland
EFA egg farmer awards
The research reported in this paper was funded by the AgriFutures Chicken Meat Program – PRJ- 010646.
NOMINATIONS are now open for the 2022 Egg Farmers of Austral- ia industry recognition awards.
them must be members of Egg Farmers of Australia. The awards will be pre-
– all work very hard to maintain a consistent and clean supply of fresh eggs for Australian families, and the awards aim to recognise this work,” Ms Hashimoto said.
That 1 percent of ana- lysed data has been the onlydriverofinnovation and insights into what we now know as ‘big data’.
This exponen- tial growth has been achievedinnosmallpart by the development of an industry extremely fo- cussed on measurement.
In 2010, the impact of recovery from the financial crisis was to drive businesses to seek software as a service solutions instead of the capital-intensive costs of running private infra- structure.
To get the machine to do what the program- mer wants, the artificial intelligence gets either rewards or penalties for the actions it performs.
This year the awards are being sponsored by Specialised Breeders Australia.
sented in two categories: • Egg Farmers of Aus- tralia Young Egg Industry
These tools can be ap- plied to big data sets to rapidly analyse data and forecast trends and con- sequences.
The awards are open to anyone who works in the commercial egg farming sector – either directly on a farm or in any support industries, such as re- search, hatcheries, veteri- nary care, transport and the like.
vice to the egg industry. Egg Farmers of Aus- tralia chief executive of- ficer Melinda Hashimoto urged EFA members to enter worthy nomina-
Last year’s winners were Dr Jodi Courtice from Queensland and Franko Pirovic from NSW.
The poultry industry has come a long way over the past fifty years in terms of volumes, cost of production, afford- ability, food safety and welfare.
The data journey
The COVID-19 pan- demic gave this extra impetus and reinforced the use of SaaS solutions, with the need to connect people from remote loca- tions in order for them to continue to do their jobs.
One such example is the improvement in ani- mal liveweight predic- tions to the processing plant using in-house plat- form weight scales.
The only criteria is that both the candidates and the people who nominate
and the people they em- ploy – such as farm staff, hatchery crews and vets
What does the next decade hold and how can we harness the power of data and new tools such as artificial intelligence to drive the next level of improvements and assist with the move towards net zero and sustainabil- ity?
However, there is still much of our industry today still using spread- sheets and manual re- cords!
Previously, weights were assessed to be val- id if each weight met a threshold programmed into the scale based on an expected weight for age or standard.
Achiever of the Year
• Egg Farmers of Aus- tralia Industry Leader Ex- cellence Award – for ser-
She said candidates who did not win from last year are also invited to be re- nominated.
The pay as you go model was here to stay!
As this has been de- veloped, it is now be- ing used in the poultry industry to improve the accuracy of production performance.
tions.
“Aussie egg farmers
The judges this year are Franko Pirovic, Jodi Courtice and Eugene Vil- joen of SBA, our sponsor of the awards and plati- num sponsor.
This has been a hugely successful period for the global industry and much has been achieved.
From paper records and manual graphs to excel spreadsheets and AS400 capability, the journey is a long one.
Coinciding with this timeline has been the commercial develop- ment and use of artificial intelligence tools such as machine learning and deep reinforcement learning.
FOR SALE
2015 Sanitouch Egg Processing Unit $69,300
• Model 10-CSG – candles, sanitises and grades 3600-5400 eggs per hour
• Ideal for small or medium-sized egg production systems
• Located Mackay, Queensland
Contact Rob Bauman 0429 380 191
Page 10 – National Poultry Newspaper, August 2022
Let’s look at where we have come from
Now the industry has the opportunity, with correctly formatted data sets, to leverage the base data in many different ways to help drive the business and efficien- cies.
Machine learning is a branch of artificial in- telligence and computer science which focusses on the use of data and algorithms to imitate the way that humans learn, gradually improving its accuracy.
With machine learn- ing, the threshold is now based on the real weights of the current birds.
Machine vision for flock monitoring
WEIGHT measurement and health monitoring of flocks in commercial broiler sheds involve manual handling and in- spection of birds, which is labour intensive and can be disruptive to the flock.
successful for predicting flocks with poor gait score and high incidence of hock burn and footpad derma- titis.
Novel machine vision al- gorithms have been devel- oped using a low-cost mo- nocular video camera with low camera setup and com- putational power require- ments to non-intrusively relate image-based body measurements to weights from weighing scales.
A novel machine vision technique achieved auto- mated behaviour classifi- cation for individual birds with an average of 78 per- cent accuracy for multiple behaviours including eat- ing, pecking and resting.
weight estimation and be- haviour classification in commercial shed environ- ments.
A recently completed project for the AgriFutures Chicken Meat program developed and evaluated proof-of-concept ma- chine vision techniques for performing automated monitoring of flock mo- tion, weight estimation and behaviour classification for the purpose of com- plementing human assess- ments of flock health and growth.
However, each new shed required calibration of multiple flocks and the behavioural correlates of the video analysis were un- known – that is, flock mo- tion was measured in terms of image pixel values rath- er than bird behaviours.
A root mean square error within 5 percent of manual measurements provided by the farm was achieved.
No other examples of non-contact behaviour sensing for commercial en- vironments were found in the research literature, with automated behaviour clas- sification typically involv- ing tagging animals with radio frequency identifica- tion tags or accelerometers for research purposes.
Commercialisation op- portunities are currently being explored to pursue commercial development of the proof-of-concept machine vision technology and to perform additional on-farm evaluations.
It is anticipated that ma- chine vision monitoring will provide early alerts to chicken farmers of adverse conditions, thereby allow- ing management interven- tion, such as checking litter quality and temperature conditions or follow up veterinary inspection. Flock motion
It is expected that auto- matedmonitoringofflock behavioural attributes would enable prediction of health with reduced num- ber of flocks required for calibration, potentially en- abling greater ease of use for new sheds.
However, research was typicallyperformedinpen experiments or semi-con- trolled shed environments, with different weight esti- mation models being de- veloped for different stud- ies.
The authors are grateful to the project steering com- mittee and to the farms where video data was col- lected, and to the Univer- sity of New England for ac- cess to experimental pens for initial camera tests. Cheryl McCarthy
and Derek Long University of Southern Queensland
Previous literature tech- niques for video analysis of broiler flock motion have been reported to be
Broiler weight monitor- ing typically involves a sample of birds from the flock being placed on weighing scales each week.
Chicken behaviour pro- vides indication of health and welfare, yet presently typically requires human visual inspection to clas- sify and quantify.
Newly developed video analysis algorithms have established flock behav- ioural attributes that are highly correlated – R2 0.7 to 0.9 – with previous pixel-based flock motion indicators of flock health.
In the research literature, numerous studies have demonstrated models for estimating broiler weight using machine vision that detects specific body size parameters from top view images of birds.
Potentially, the developed machine vision approach could be used for develop- ment of objective welfare metricsbasedonbehaviour quantification or for input into a climate controller inside commercial shed- housing systems. Conclusions and further work
Machine vision has appli- cation to monitoring tasks for other housing systems and this is presently being developed in projects for the chicken meat and egg industries. Acknowledgements
Weight estimation
Behaviour classification
Novel machine vision analysis of video streams from low-cost colour cam- eras was demonstrated to be capable of simultane- ously performing real-time flock motion monitoring,
Improved poultry production performance
n Computer learning on animal health and welfare
MANY would be shocked to know that researchers analyse and gather insights from only 1 percent of the world’s data.
achieved because of the genetics companies ap- plying technology and powerful data systems in their breeding pro- grammes.
Facebook, Webex, Dropbox, Google Drive and iCloud were first made available in the early 2000s and we all became more comfort- able with these types of solutions.
Reinforcement learn- ing is the training of ma- chine learning models to make a sequence of decisions.
The other 99 percent of the one quintillion bytes of data that is collected every day – according to a recent study from International Data Cor- poration – remains un- touched.
Measurement of key performance indicators, a great focus on improv- ing the genetics we are working with and a will- ingness to pass on some of the efficiency bene- fits to the consumer are making poultry products more affordable and in turn driving more vol- umes.
Its goal is to maximise the total reward.
Over the past 60 years, the global poultry indus- try – both meat and table egg sectors – volumes have grown massively, and the genetic poten- tial of the broiler and the egg layer now bears no resemblance to what was achieved decades ago.
Computing capacity is now cheaper and more available to everybody than ever before. Artificial intelligence
Machine learning is an important component of the growing field of data science using algorithms to build a model based on sample data – known as training data – in order to make predictions or decisions without being explicitly programmed to do so.
As weights are re- corded, they are then assessed, analysed and used to create a rolling set of average weights.
In fact, in the past 10 years we have seen considerable improve- ment in performance – not only in growth rate, FCR and egg numbers but also in welfare traits such as gait, mortality and feather cover.
In recent years there has been a new tool to add to our ongoing search for ways to crunch the mass amount of data the industry generates.
Not only does this make the weights re- corded more accurate as the thresholds can be smaller ranges but we can also use this constant rolling number to accu- rately predict the coming weights of these birds creating more confident weights for the plants. Animal health and welfare examples of positive use
Much of this is
The opportunity to move data into the cloud has unlocked consider- able additional data pro- cessing capability and speed.
Great strides have
Step forward artificial intelligence.
This is the main dif- ference compared to the mathematical models that can easily get stuck without giving any an- swer if not tuned and balanced properly.
* continued P11 www.poultrynews.com.au