Jimmy Carroll: [00:00:00] Hi everybody. My name is Jimmy Carroll. I’m the vice president of operations at Tech B2B Marketing. I have the pleasure of being joined today by Jeff Oleske of Specim and obviously my colleague John Lewis at SPIE Photonics West 2024. Jeff, thanks so much for taking the time.

Jeff Oleske: [00:00:14] Yeah. Thank you. Good to see you.

John Lewis: [00:00:19] Jeff, for those that don’t know much about Specim in our audience, could you talk a little bit about the company?

Jeff Oleske: [00:00:24] Yeah, sure. So Specim, Spectral Imaging, it’s based in Oulu, Finland, that’s where we do all our manufacturing. We started in 1995, so we’ve grown to the largest commercial hyperspectral imaging company in the world. We’ve got about 90 employees based globally. We’re in the U.S. We’ve got a team here in the U.S. We’re based in China, we’re in Spain and Germany. So yeah, we’ve really grown a global presence. And we manufacture hyperspectral imaging cameras for a whole range of applications, starting with agriculture and airborne studies back in the ’90s. We’ve grown into really the leader in industrial hyperspectral applications, ranging from the visible to the near infrared all the way out to the longwave IR.

John Lewis: [00:01:09] That’s interesting. Could you talk a little bit about how you define hyperspectral imaging?

Jimmy Carroll: [00:01:17] Especially in the context of multispectral versus hyperspectral.

Jeff Oleske: [00:01:23] Yeah, it’s weird to have that definition because some will define 10 spectral bands as hyperspectral. We would see that more as multispectral. We have over 200 bands. So we’ve got 224 bands in some of our cameras. Certainly we look at hyperspectral as hundreds of bands. Multispectral is another technique. It’s just less bands. So where you draw the line to differentiate hyper- and multi-, it’s a bit blurred. But certainly when you’re into the hundreds of bands, we consider that hyperspectral. And it’s a continuous band, a continuous range of wavelength bands, across all of our cameras. So, for instance, if you’re in the visible to near infrared, we’re covering 400 nanometers out to 1,000 nanometers continuously. We utilize a push broom line scan technology, which is very much like your traditional spectroscopy, using a slit, looking at a single point at a time, and then seeing the entire spectral range for that spatial point. But then imagine having a thousand of those pixels across, which is what gives you the spatial component. So we have both spatial and spectral dimension, all in real time, all co-registered with one another. So every spatial pixel contains the full spectral range. We don’t use filters. We don’t look at pixel-to-pixel variation. It’s all done concurrently.

Jimmy Carroll: [00:02:55] Hyperspectral imaging in general, it’s always been — I shouldn’t say always — but in the last let’s say six to eight years, it’s been a really hot topic in the industrial space. And it’s evolved from sort of this large lab-based project to something that’s — your cameras are fairly small now, right? I don’t know if it’s this big or even smaller.

Jeff Oleske: [00:03:15] Yeah. About that, yeah.

Jimmy Carroll: [00:03:16] In general, talk about a little bit the state of hyperspectral imaging today in the context of manufacturing and industrial applications.

Jeff Oleske: [00:03:25] Yeah. I mean, when we talk about the ’90s, even the early 2000s, when Specim was was a bit younger, the technology was more expensive, right? It was less attainable for industrial customers to get ahold of. It didn’t make sense financially. Now the camera technology has advanced, and with that has come competition, which is actually great. It drives the price down. It allows more industries and more applications to take advantage of the hyperspectral technology. What we have done, especially over that time frame you mentioned, these eight or so years, has been a real heavy, methodical push into the industrial world to drive down not only the overall cost of the technology but selling in volume to allow us to manufacture for less. So while we’re focused on the industrial applications, we still have our foothold in the research market. So we’ve generated basically a market of using research-grade spectroscopy cameras that are hyperspectral in nature and applied them to the industry, which has done a few things for the back and forth, to benefit both sides of that coin, you have technology and advancements we’ve made for the industry, like a unified calibration.

Jeff Oleske: [00:04:48] All our cameras are unified, calibrated, so you can plug and play,  in the field. You can hot swap. Well that’s really nice for the research market too. You have collaborators that may be global. You have one lab working on one project. You can use the same model in another lab somewhere else without having to change your calibration. So innovations like that have really pushed the technology forward and made it just more adaptable to the industrial use.

John Lewis: [00:05:15] Now, is that calibration, was that driven by going into the industrial market?

Jeff Oleske: [00:05:19] Yeah. That was something they needed. They had to have this unified calibration because they were either growing and going on multiple lines. They needed to have a hub basically of R&D, and that camera needed to be the same model that they were using out in the field. But now when you apply it to the academics, yeah, they like it too. It’s good when both sides kind of drive the technology forward.

John Lewis: [00:05:43] Well, speaking of the industrial market, I’m wondering if you could talk a little bit about the key applications and segments of the industrial market where you’re seeing hyperspectral imaging.

Jimmy Carroll: [00:05:53] It’s got to be food and beverage.

Jeff Oleske: [00:05:55] Food for sure. Food and beverage, foreign object detection, quality control, basically everything that the human eye can’t see, things that RGB can’t see. So when we’re out in the field and we’re talking to people trying to solve their problems, they’re either using humans to inspect and they want to go further than that or they’re already using maybe a line scan RGB camera, which has its place, and we don’t compete with that. But what we do say is when you need more than just red, green, and blue, when you need the 200 bands or even 50 bands, that’s where hyperspectral can come in. And now we can see things that just traditional vision can’t. So we’re solving problems that in the past have just been chalked up to “This can’t be done.” And now we can accomplish those. Other industries: we’re really big in recycling. So sorting plastics from other materials, construction waste. Textile recycling has actually become a really hot topic. There’s companies out there vowing to be zero waste and circular economy and all this, all these hot topics. But in order to do that, you have to identify what the material is. You have to sort it. And when you have a pile of blue shirts, being able to sort those by nylon and cotton and polyester, it’s really important, and spectrally it’s easy to do. There’s other industries too. We’re in the mining industry, and our airborne agriculture market is still very big, remote sensing. We just launched a new longwave infrared camera. So the FX120 is brand new. That’s really meant for the mining industry, looking for minerals, core logging, core scanning, but also airborne, where you can look for thermal imaging. There’s a lot of applications for that as well.

John Lewis: [00:07:46] I’ve noticed that Specim has had a few webinars out in the last year, one focused on NIR, near infrared applications and another one on shortwave infrareds. And I’m wondering, could you talk a little bit about those different bands of the spectrum and what drove you to develop those kind of products to address those certain bands?

Jeff Oleske: [00:08:10] They’re all market driven. They’re customer driven. So we have the expertise to make a hyperspectral camera. It’s by changing the optics, by changing the detector, adding the cooler, adding the other components, you can now explore different wavelengths. The core competency is there, of developing these high-quality hyperspectral cameras. So starting with the visible and the near infrared and then launching a shortwave camera, it just made sense to find applications to keep going.

Jeff Oleske: [00:08:39] The FX50 is our mid-wave camera. It was developed, and now its main application area is the black plastic sorting. So sorting black plastics has been an impossible feat. There’s actually people that think you can’t sort black plastics, but you can with mid-wave IR, hyperspectral you can. You can Identify material based on their chemical signature, even if they’re black. So to the human eye or even into the near infrared, they can be carbon black and absorb all of the light. But in mid-wave it doesn’t. There’s reflection there and there’s signal there, which again will help with this circular economy. Effectively, the recycling industry is taking recycled plastics and dyeing it black so that they can use it and cover up all the impurities and use it in,  you know, our laptops and our electronics, all of that’s black plastic. But then what do you do when you’re done with that material? That now needs to be sorted. And this is a way to do that.

John Lewis: [00:09:37] There’s a lot of black plastic under the hood of a car too.

Jeff Oleske: [00:09:40] Look around you. There’s black plastic everywhere in this room right now. So yeah, there’s black plastics everywhere. And this is a way to sort that and be more environmentally conscious.

Jimmy Carroll: [00:09:55] Interesting. One thing I wanted to ask about, Jeff, is on the software side. And it kind of ties into this trend that I’ve heard a lot on not just this podcast but in the industry everywhere, in LinkedIn and people’s marketing and everything, is the idea of ease of use. And obviously people want their products to be fairly easy to use. But the ease of use in the industrial space typically will come from the software side. So how does Specim approach that?

Jeff Oleske: [00:10:24] So software drives everything. I mean, the hardware can be great. It you don’t interface with software, there’s only so far you can go.

Jimmy Carroll: [00:10:30] It’s just a cool-looking camera.

Jeff Oleske: [00:10:32] I mean you need adoption, right? You need people to buy in. You need it to be easy, and you need it to integrate with things that they’re used to. So when we’re talking about the industrial world and machine vision, it’s a very well-established market. There’s well-established cameras in that field already. To get into with hyperspectral, we need the software and we need to make it as easy as possible. And that’s actually our tagline is to make spectral imaging easy. Maybe we’re making it easier. Maybe that’s the real way to phrase it. We’re making it as easy as possible. Hyperspectral imaging is not an easy thing. It’s data intensive. There’s a lot of data, but we’ve made it as easy as possible to use our home-built software to take the raw data and analyze it and process it and give you a processed hyperspectral image that’s effectively color coding each pixel. So if you have different classifications of material and you’re trying to sort it in real time on a conveyor, you can take something that visibly to the eye looks identical. Take pill sorting, for example. You have two different types of white pills. They’re the same size, shape, color. No way a visible system would be able to differentiate. But because they’re chemically different, they have a different absorption signature and you can sort them. You can see in real time the spectral signature of those and code each pixel. This is a blue pixel. This is a green pixel. And now you can have a robot or some kind of picker or automated system then take that processed image, which is now much lower data intensity. It’s a much, much smaller file. It’s now just an RGB image created from this hyperspectral cube, and then you take action on that. So we’ve made it as easy as we possibly can to really plug and play with our SpecimONE platform, which combines the camera, the processing computer, which is called the cube, and then the INSIGHT software Specim INSIGHT software, which will do the analysis. So putting all that together, you can output an RGB image in real time based on hyperspectral data that you have to process quickly in order to get a result and then have robotics or some other form of picking and choosing what to sort.

John Lewis: [00:12:56] You talked about ease of use and plug and play several times. And from how you describe the system, a lot of companies may sell a hyperspectral camera, but it’s kind of up to the end user to figure out: What do I do with this? How do I get the software and the processing to all tie together?

Jeff Oleske: [00:13:14] Yeah, for sure. There’s lots of software out there. Like many hyperspectral companies, we interface with all of these different analysis softwares. The difference is we’ve created the SpecimINSIGHT ourselves. So it’s kind of a home-built thing to do the classification. But the data is also a raw data cube. If you want to take that and work in MATLAB or Python or code your own language, you can do that as well. And there’s industries out there that want to do that. They want to have a more customized feel, and that’s fine. We work with that as well. But again, the idea is to make it as easy as possible, get people up and running as quickly as possible, and solve problems that traditionally aren’t solved. And really getting the word out that this technology can potentially solve your problem. So if you have an issue, if you have an application that you’re struggling with, hyperspectral is usually a solution that you should consider.

John Lewis: [00:14:10] So just off of that, so what would that look like? Like if I have a problem, I reach out to you. Do I have to buy a camera? Can I send you stuff, samples for you to look at?

Jeff Oleske: [00:14:26] We are very versatile. We’d like to help you any way we can. We’ve done feasibility studies, where we can run it in our lab and show you that the feasibility of it is true, that we can classify these materials. Whether or not that works in real time or under your conditions, that’s where we have to start talking. But it all starts with: Is it spectrally different? If the answer is no, it’s not spectrally different in the visible, the near IR, the shortwave, the mid-wave, the longwave, we can try all our cameras, and these they’re not spectrally different. Well, hyperspectral is not going to be a solution. So once we get over that hump of, Yeah, we can classify these, they’re spectrally different from one another, then we can talk about the use case, you know, what’s the frame rate required? How many spectral bands do we need? Because that’s the other thing. We have these 200 bands, but you may not need 200 bands. So our hyperspectral cameras can effectively become multispectral. You can cut down the amount of data that you have to process. You can speed things up. You speed up the frame rate of the camera. So we’re pretty versatile in terms of meeting the needs and the priority list from a customer. If speed’s your main priority, then we can start there and we can work our way backwards. So there’s a lot of ways to solve the problem, but that’s one where we could definitely do a feasibility . . . we do on-site evaluations. We work with partners that can integrate full solutions. We’re really the hardware provider. We do the hardware and the software to do the analysis. But we don’t do the robotic integration. We don’t do that. The conveyor integration. We rely on partners to help us with that.

John Lewis: [00:16:07] Sure. That makes sense. Yeah.

Jimmy Carroll: [00:16:09] Jeff, not just for Specim, but what’s the future hold for hyperspectral imaging hyperspectral cameras? Like, where could you see it being used in the future?

Jeff Oleske: [00:16:19] Every industry that we’re in now started somewhere, right? So the mining industry started as kind of core logging and, you know, take a drill core and see what minerals are there and then make a decision. By getting more and more real time and expanding into the mid-wave and the longwave, we’re just uncovering more data that wasn’t available before. So even within the applications that we’re currently in, we’re seeing kind of new branches starting to sprout from that. We listen to the market. So if you’ve got an idea or an application that you haven’t been able to solve, we can see if it works. You know, we can’t come up with everything, but we’re in seal inspection. We can look at the seal integrity of a package, pharmaceuticals, food. So really everything is a chemical. Everything has a signature. It’s just a matter of which wavelength they absorb. So once we identify the wavelength, then, yeah, we can solve almost any problem that you can come up with.

Jimmy Carroll: [00:17:30] Anything else we haven’t talked about, John, that you want to ask or Jeff that we haven’t brought up?

John Lewis: [00:17:36] I was just thinking, our last guest was talking about going from a 3 megapixel smart camera to a 5 to a 15 and how you’re getting more and more data. People want more data. But then there’s the trade-off of, you have to handle all this. Where do you put it? And I was wondering if you can just comment on . . . because hyperspectral, that’s about as much data as you can get.

Jeff Oleske: [00:18:01] There’s a huge amount of data, but it’s not necessarily — you don’t need to store it in that form. You don’t need the full data cube if you’re doing real-time inspection. You can analyze it, pump out an RGB image in real time, and then you don’t care about that data cube anymore. You’ve already got your answer. In research, you want to collect the whole data cube. So then you have to talk about, well, what’s the right amount of pixels, how much data can I store, and then go from there. Are you using a cloud-based system to do the analysis? That’s becoming more and more popular. We’re getting asked about that more and more. Can I stream here and then stream to a cloud and then have colleagues or collaborators take that data and do the analysis on different sites? Sure. But it’s a lot of data. So yeah, you either have to package it or clean it up in some way. So it’s interesting, but I’m not sure what the future holds for that kind of stuff. Obviously with AI and all now, it’s becoming a competitive marketplace of wanting more data. So they want more data to train more tools. And that’s a good thing. So we’ll see what happens with that. But yeah, it’s a lot of data. But again, when you’re doing the real-time inspection, it doesn’t have to be. We can do that processing in real time and throw it away.

Jimmy Carroll: [00:19:27] Jeff, if people want to learn more, is it specim.com?

Jeff Oleske: [00:19:30] Specim.com. We’re also on all the social media platforms. Again, we’re global, so yeah, look up the “contact us” field on the website. And yeah, if you’re based outside of the U.S., we have people that you can talk to. But within the U.S., you can go through me and my team. And we’re happy to talk to you and figure out what works, where hyperspectral should fit in if it meets your needs. You know, it’s more about talking about new applications and identifying where we fit and trying to find, like you guys are talking about, find that new thing that we can do, so if you’ve got a problem that traditional vision can’t solve, hyperspectral is very likely a solution. So something to look into.

Jimmy Carroll: [00:20:18] And, you know, if you’re at an event, trade show, conference, a networking event, you can find Jeff. He’s really difficult to find.

Jeff Oleske: [00:20:25] I try to stay highlighted.

Jimmy Carroll: [00:20:26] Yeah. Well, Jeff, thanks so much for taking the time. Really appreciate it. Again, this is the Manufacturing Matters podcast, manufacturing-matters.com. If you have any questions, comments, want to reach out. If you have questions for Jeff, feel free to reach out to us and we’ll pass them along to him. So thanks everyone for watching.