Design-build and consulting company Wiley recently partnered with Meat and Livestock Australia (MLA), the marketing, research and development body for Australia’s red meat and livestock industry, to explore how augmented reality (AR) could be utilised to simplify the country’s meat grading process. Wiley R&D and innovation director Brett Wiskar explains how the platform works.

The grading of meat and carcases is a vital part of the production chain, maintaining quality levels and providing an important line of defence against harmful meat entering into the supply chain. But despite its importance, meat grading in Australia is still largely a human-led process, with individual workers being called upon to make subjective decisions on carcass and meat grades.

Elliot Gardner: Tell me more about the augmented reality (AR) meat grading project

Brett Wiskar: As part of Meat and Livestock Australia (MLA)’s ‘Precision to Decision’ investment programme, we took a look at the grading process in red meat factories in Australia, called Meat Standards Australia (MSA) grading. In 92 different processing facilities across the country, there’re around 200 workers who measure attributes of livestock carcass when they’re on the hook. These individuals will measure along 11 metrics, including temperature, pH, etc.

Currently, the industry relies on a subjective means of capturing that information. In other worlds those 200 or so individuals all get tired or have off-days, but the industry relies on them making really accurate decisions consistently, so there’s inevitably going to be a factor of inaccuracy in that process.

We talked to the MLA and said this is an area that a piece of computer vision software and potentially some sort of AR interface would be able to assist with. Call it a decision support system; or perhaps a path to objectivity.

We built a prototype, or proof of concept, so we could prove the software would help in this decision-making process, and lead to a more consistent set of outcomes across the industry.

EG: Do you want to remove subjectivity entirely from the process?

BW: If that was an option! If you could move to a completely objective process I am sure that everyone in the industry would be on-board with that.

We live in a world where humans make decisions. Because of the nature of different processing facilities and the different kinds of animals being graded, as well as a range of other factors, it’s not feasible in the short term to remove the human from the process.

Humans are still very useful for moving cumbersome things around, for holding a side of beef open when measurements are taken, and for lots of other things that it’s very difficult to design an automated process to do.

Instead of removing the human, it’s more about augmenting a person and the decision making process the human makes.

So I wouldn’t necessarily say we’re on a short-term journey to remove subjectivity, what we’d like to do is reduce it and move towards some degree of objective measurement, so it doesn’t matter which grading facility you send your animal to in the country, or which grader you encounter on a specific day.

EG: In terms of the interface, what do workers see?

BW: So some things, like temperature and ph are actually graded with instruments at this point of time, but there are other areas, such as measuring the square centimetres of the latissimus dorsi muscle, that AR can help with.
Traditionally graders hold up a little transparent acetate grid over the meat and actually count the square centimetres. That process is time-consuming and not necessarily accurate all of the time, so what we did was build a piece of software that can build up a reference point for the computer vision system to look at, giving the software the capability to determine where it is in three dimensions versus the cut of the meat. Essentially a grid is generated in the AR interface that calculates the number of square centimetres there are in the red part of the muscle.

When the computer vision detects a transition from red to white as we get to outside of the muscle and into the fat, then it understands that that’s a contour and the edge of the meat, so the edge of the area that needs a grade.

You can also point a camera at the meat to determine colour, instead of the current method of a human holding a colour sheet up against it. Not only is that process somewhat laborious, after you’ve done 700 hundred in a day, all the colours start to look the same.

It’ll also be able to scan and read the barcode that’s on the carcass tag, and relay it to the user in real-time.

EG: What’s the future of the project?

BW: The version we’re created is very much centred on MSA grading. That said, there are areas related to disease control and detection in other foods we could expand to.

Even within red meat, there are other applications to consider. In abattoirs, for example, viscera from animals gets taken to a vet to inspect. Computer vision and artificial intelligence could be used to look at whether there’s a high likelihood of a specific piece of meat having a particular condition, and could highlight to the vet that they need to keep an eye on it.

With our project though, we’re taking a journey over the next four or five months to go and work with the largest processors in Australia to show them what we’ve created, to help them understand that this is just the first step on the road towards a commercialise-able product that would increase the precision of the industry.

EG: With Australia being one of the world’s largest meat exporters, do you think this technology would be able to set a new standard?

BW: It’s an interesting question. Every red meat market in the world has a different means of grading. We’re certainly not going to see markets around the world standardise their grading processes. Most countries closely guard their system and suggest that theirs is the best in the world and that’s why theirs should be trusted over other countries.

What you would find, though, is that the MSA grading areas we look for on the AR platform are similar to that would be looked at in the US or Brazilian systems, or any other beef market around the world. The software could therefore be modified to work with their standard. That’s likely to be a remarkably short journey.

That doesn’t get you through all the regulatory compliance and testing required to demonstrate that this is a suitable solution, but the software itself wouldn’t take an age to write if you’d already achieved commercialisation in one market.