Behind the Scenes: Real-Time Lean Point Measurement: How PPO Transforms Meat Quality Analysis

 Video Transcript

Welcome to Behind the Scenes with PPO, a video series that offers a look behind the curtain at how meat processors collaborate with tech companies like PPO and why.

I’m Heather Galt, the Chief Customer Officer at P&P Optica.

And today, we’re speaking with Tim Stork, our senior solutions engineer here at PPO.

And we’re going to chat about how PPO helps meat processors go beyond foreign materials detection to measure product quality in real time.

One of the interesting things about Tim is he’s one of our longest standing employees here and has seen the company right from its early stages through all the pieces of our development to get us where we are today. So welcome, Tim. Thank you. So, Tim, you’ve got particularly unique expertise in this area because you’ve seen the evolution of PPO from the very beginning. And, you know, we started in foreign materials, but as we’ve grown, we’ve evolved into being able to measure product quality across a number of different measures, things like lean point. Can you give me a little bit of background on how we got there and why PPO is uniquely positioned to do those product quality measures?

Absolutely. So, I think we all know at this point, PPO is world class for material detection, and one of the things that makes us so special is our hyperspectral imaging, the near infrared, all that kind of stuff. So, that is a very natural fit, not just with foreign material detection, but the amount of information in that spectral data can and and has been proven to be able to be used for so many things in the meat industry. And one of the big ones is lean point. This technology has already been used for decades in labs to do exactly this application. So, it is really a natural fit with our expertise now in being able to scale up and have our machines work in, the food processing setting that we can add on applications like lean point, but also things like, meet quality and, and other vision applications and expand the value that we can bring our customers.

Challenges in Real-Time Applications

So you mentioned a really important point, which is that HyperSpec has been used in labs for a long time. So we don’t operate in a lab environment. Right? We’re online, real time, in operation. So what has PPO done differently so that we can actually use HyperSpec in line for lean point? Yeah.

So that leap was, I think, the reason why PPO is still the only one in the meat industry being able to use this technology.

There’s a few challenges. One of them has to do with the sheer quantity of data that is hyperspectral imaging. So something like terabytes in a shift, if you were just to try to catch all that information, which, isn’t really feasible. So Right. Two things have made that kind of work. We’ve got, a, computers, are pretty widely available with, with a lot of power and in a small package. So we take full advantage of that. And then the analysis side, I think, is most critical.

So the ability to take all that information and not try to save it, but we don’t have the luxury of time in a lab where you let the machine run for ten minutes and it spits out an answer. We need an answer in a fraction of a second Yep. To make those live decisions or to help with automation in the plant. Right?

So, machine learning is, and AI are the only ways to do that. And we have so much learning from foreign material detection where we are training these models to make decisions very quickly. We can reject things in a fraction of a second. And we took all that learning and applied it to these other applications where, yeah, we still need that information within seconds in order to to, make decisions and take action. So it just keeps building on on, again, our, our expertise, what we do that’s that’s really unique. Got it.

Comparing Technologies: PPO vs. X-ray

And so with Hyperspec, you know, we’re we are a little bit of a different technology than, say, an X-ray or some of the other methods that are used to measure lean points.

So if I’m a meat processor, what should I expect from PPO’s version of the solution versus what I might be using today, whether that’s in a lab or maybe an X-ray or something else?

Yeah. So over time, there have been different ways to do this. Of course, lean point is still critical in these plants, for a number of reasons, but they need to know what’s going into their product. They need to know that their product is meeting all the specs. Right?

So back in the day, you used to have a person who had worked with meat and handled it for thirty years, and they would look at it and stick their hand in and say, yep. That meets what this product is supposed to be. That’s not really a sustainable future for these plants because they have turnover and these experts are are retiring. Right?

And there’s not the same interest to become a lifelong meat expert that people set off in the beginning of their career. Right? Yeah. So, there’s been a number of technologies.

The Limitations of Lab-Based Approaches

I’d say the one that’s now starting to get phased out is that lab based approach. It’s relatively cheap. You buy a benchtop piece of equipment. You have it in a lab.

You take a small sample of meat off the production line. You prep it. You put it in the machine, and you wait. So that technique, I’d say, is about a turnaround of maybe fifteen minutes from when you take it off the line till you get an answer.

Of course, all of these things have trade offs. And for the lab based one, it really is well, I’m taking a small sample from thousands of pounds of meat. Right.

I put my hand in. Yeah. Yep.

What’s to say that that handful of meat or lots of little bits of meat are really representative? And that can add way more error than the equipment itself. The other thing is just waiting fifteen minutes has a cost. So if you’re making a batch of meat, either you’re saying I need to make sure it’s good, and you’re waiting for fifteen minutes, which is like, what factory wants to wait for anything?

How much meat can they process in that quarter hour? Right? Yeah. Or else you’re saying, I think this is gonna be fine. Let’s act like it is. And if we have a problem, well, then we need to deal with it.

And that’s a huge pain too because now you’re saying, well, hold the phone, pull back fifteen fifteen minutes could be thousands and thousands of pounds, and, that’s a big pain.

Absolutely. Absolutely.

Now the other, so PPO’s real, wow factor here is that we’re operating online in real time. So we’re not the lab based. We are, we are looking at all the meat as it goes by and providing real time information. Yep. So, fifteen minutes, not a thing anymore.

Yep.

And also the is that tiny sample representative not a thing anymore? So, really big benefits there. Now there are other technologies out there that can go on the line and look at all the product. And the big one is X-ray systems.

X-ray Technology Explained

Yep. So these systems have been around at least a decade, maybe a little bit longer, and it uses x rays, usually two different energies of x rays to be able to look at the density of the product. Okay. Now there is some training and some math involved there because really the only information that you’re getting there is how dense is that part of the meat.

So, of course, fat and lean are different, but there are quite a few assumptions there that they need to do in their math based on things like moisture content. Well, fat and lean are two different things. But within lean, you may have pork that has more moisture or less.

And that X-ray technology can’t really differentiate, so they have to make assumptions.

And, And, really, that does have an accuracy impact in that you may get pigs one week that just are naturally different than another week, and those assumptions might not always be correct.

So PPO, the way that it works, we we are looking at a pixel by pixel analysis of the product as it goes by. But instead of having that kind of limited density information, we’re looking at the chemical composition. So, of course, fat looks different than protein and moisture. We put a lot of training into these systems to make sure they know to what to recognize and what that means. Yep.

Training and Calibration for Accuracy

And then with all that information, we can calibrate it to be really, really accurate.

Tim, it sounds like, you know, what PPO is doing with hyperspectral imaging does require some training and some, you know, learning for the system. I I’ve talked with both James and Keisha at different times about our AI machine learning. So, you know, if you have an X-ray, I’m gonna guess that if it’s only based on density, the adjustments that they can do in the math are probably relatively simple, and so maybe it can go online pretty quickly. What does it mean for PPO to get ready to do lean points in a plant?

That’s a that’s a great question and something that PPO spends a lot of time thinking about because, of course, we want our customers to be up and running as quickly as possible. So when we go about creating a new product like this, that’s one of the lenses. Right? There’s accuracy. There’s speed of deployment. And maybe that’s not the first thing we think about, but it’s definitely the second. Yep.

So when it comes to, how we help to, get these things running as quickly as possible, we always wanna start with some base models. And when we develop a new product, building these base models is where a lot of that early work happens. We have a lab here that’s set up like a meat processing, environment, and we do have to collect the training data and the testing data and find, a gold standard measurement technique to compare against, and we send our samples out to labs and all that kind of stuff.

And that doesn’t get us all the way there. But having the core model for pork and beef, for example, because that’s what we’re deploying today, makes really speeds things up. Now when it comes to the final deployment, we know no two plants are the same. Just like in foreign material detection, we need to make sure that we’re gonna be as accurate as we can be in that plant.

And hyperspectral imaging is able to see so much nuanced information. So what cuts of meat are you running? Where did those animals come from? How have they been stored?

Right? Was that meat frozen or fresh or thawed? What’s the moisture content? So we need to optimize those models in the end to make sure that it can recognize those differences.

And if something has been frozen, that doesn’t throw off the the measurement accuracy. Right?

So that last optimization is what we do in the commissioning process. We collect some more data in the plant on the real product.

We send off for some more, lab tests to confirm and calibrate that final accuracy. So we can basically show these reports to the customer and say, yeah. Here’s your proof. We have met what we said we were gonna do. Look how awesome our systems are running. Got it.

The other thing I think x rays sometimes struggle with is foreign finding foreign materials and doing lean point at the same time.

Doesn’t always seem to work, I don’t think. At least that’s a feedback I hear when I’m out in plants myself.

So how does PPO not have that problem? Because, you know, what I’ve seen in plants is we can do both at the same time. Yeah. But I don’t know how to explain how we do that.

Yeah. Really good question. So, absolutely, one of the biggest benefits of PPO’s technology is that we can run multiple applications at the same time making all those decisions, and it really doesn’t have any trade offs. So, yes, it comes down to the computing power, but computing power isn’t really a limitation for PPO today based on what’s available.

So what an x-ray does is, again, it has a limited access to information. It’s got maybe two different energies of x rays, and there’s a bit of tuning to be to say, hey. If I want to improve the accuracy of my fat lean, I can change these things accordingly, and I can squeeze a little bit more performance out. But on the flip side, if what you’re looking for is I mean, x rays can find bone and metal and high density foreign materials.

You got to tweak it a little bit in the opposite direction, and you now have this trade-off just based on the physics constraints of how those systems work.

So basically, the two wavelengths versus the five hundred or six hundred plus that PPO can use to get the same information.

Exactly. So when we look at PPO and that hyperspectral information, we’re collecting all that information anyways. Right? Right.

That is what our spectrometers are always doing. So then it really just comes down to analysis. Are we running models to do foreign material? Yes.

Can we also run a completely separate analysis model on fat lean? Yes. So that lets us, basically, as far as we can tune the fat lean and as far as we can tune the foreign materials, you can get both of those benefits optimized in the system. Because you’re basically taking the same data Yeah. And processing it twice. Exactly.

So then looking forward, it sounds like we could take that same data and do other things with it. So what do you see coming in the future then for PPO?

Yeah. So it has been a really interesting journey over the last ten years seeing how PPO has has grown and innovated and and really been able to change the meat industry. Foreign materials was a a really great starting point. But, I mean, even so within each of our products, we’re constantly improving, getting more accurate, pushing the boundaries of performance.

And then we have this list of all these ideas of things we’ve heard from our customers and ideas we’ve had about new other applications we can do. So, we talk a lot about our hyperspectral cameras. We have a vision system in our in our system as well. So that unlocks, kind of these dual applications where we can look at the chemistry, we can look for quality, we can look at composition things, we can compare that with things like size and shape and, color and, interesting things like that.

So, we over time have done quite a number of feasibility studies on, things like tenderness of meat or, I think one of the dreams, that HyperSpec should be very capable of is to look at flavor. So can you look at a steak while it’s raw? Maybe it’s just been cut in a factory and be able to tell after cooking how will that taste to the consumer. Wow.

That’s like the, a real game changer in the industry, and nobody’s been able to do that.

Well, and that and that’s truly farm to fork, which is what we’ve always talked about. That’s pretty amazing stuff.

The other interesting, one we’ve been working on has to do with, disease detection. Right? So this is more to the farm side. But, today, across the world, there are humans who are looking at various pieces of the animal to say, is there disease here? Do we have to discard parts of this animal or maybe even the whole thing?

And anytime you’re using people, they’re gonna make mistakes. There’s gonna be they’re gonna miss things or they’re gonna, over grade some stuff, and all that has impacts on the business and on the profitability. So, PPO, again, the hyperspectral data, we can see those nuanced things. We can train models to do it, and we can automate something that’s never been able to be automated before.

And that helps give feedback to the farms. That helps people on the line do their jobs. That helps deliver better quality product to the consumer as well.

Wow. That’s awesome. Well, thank you so much, Tim, for making time for us today. This is a really informative conversation. I always learn something when I talk to you. Don’t forget to check out the other videos in our Behind the Scenes series, including our fascinating interview with Alex Heater, Technical Lead for Strategic Initiatives at PPO, who explains why there’s no such thing as a perfect detection system. You can find a link to that video in the description below.

See you next time.