ANALYSIS AND OPINION: HYPERSPECTRAL IMAGING IN MACHINE VISION Reaching next-level machine vision P
ro-Lite Technology’s Nick Barnett described spectroscopy as “measuring at one point and getting
a full spectrum, giving you a fingerprint of what materials might be in the object you’re looking at,” at a seminar on non-visible imaging at the Machine Vision Conference this year. He went on to describe other imaging methods including monochrome cameras, “which just look at the greyscale”; RGB, which gives just “red, green and blue” and multispectral, which “uses 8-12 bands in different areas within the spectral range, whereas hyperspectral imaging is really a continuous range of wavelengths with regular spacing.”
Explaining hyperspectral imaging “Hyperspectral cameras,” explained Barnett, “look at a very broad range of the electromagnetic spectrum, so users aren’t limited to working in the visible region. [Because] there’s a lot of information in the infrared, short-wave infrared and mid- and long-wave infrared, the potential for hyperspectral and multispectral [imaging] is to provide a lot more information about the materials than a colour sensor.” If you’re looking at water and oil, for example, both are fairly transparent in the visible range. But, in the shortwave infrared, it’s “easy to pick up the differences.” “With hyperspectral,” said Barnett, “you
get a hypercurbe,” with each pixel of a 2D image having a full spectrum. “So there’s a lot of information contained within.” If this sounds complex, however, there’s no need for concern, as Barnett predicted we’d all become far more familiar with hyperspectral over the coming years, with the technology being used in many future applications. “For example,” said Barnett, “you can detect clouds and methane using spectral cameras. You can look at forest fires, tree health, soil composition, algae blooms and also at urban scenes to identify different buildings and trees.”
Hyperspectral imaging in machine vision “The applications [of hyperspectral imaging in] automation and machine vision are broad,” said Barnett, with example use cases being in mining where rock
“Making sure what’s being packaged up is actually what we thing it is, is more important now than ever before”
with hyperspectral imaging Pro-Lite’s Nick Barnett explains how machine vision users can harness the power of hyperspectral imaging to move away from the conveyor belt and into real- time, 3D inspection machine vision
cores are scanned or rocks or asbestos are sorted from normal building materials, or in the food industry, pharmaceutical, waste, textile or plastic sorting, where there are a range of wide application spaces. Hyperspectral cameras can produce classified images, detailing different materials’ chemical compositions as part of quality assurance processes, or quantify materials such as the amount of water, fat or protein content in cheese.
Food and material security “You can look at contaminated scenes,” suggests Barnett, “you can identify nuts and maybe some shells and sort them out.” Because making sure that what’s being packaged up is actually what we think it is, is more important now than ever before. Meanwhile, in recycling and plastic sorting, shortwave infrared is able to sort through and classify the spectral signatures of PET and PVC.
Using the data for spectral imaging “[Once] you collect lots of data, you build a model,” said Barnett. “That might use a neural network, or spectral signatures from a library, but you can build a model for classification and quantification and run it in real-time. To improve it you can add more samples. Then you can build up a classification algorithm and finally, you can see the materials of individual samples. “Looking further into how this is used
36 IMAGING AND MACHINE VISION EUROPE AUGUST/SEPTEMBER 2024
in a real-time application,” said Barnett, it can “sort what’s PET and what is another plastic we don’t want. As the plastics label is on the conveyor belt, the spectral camera classifies each flake from the image, and uses air nozzles to fire them into different bins. It’s a very fast and efficient process.”
3D hyperspectral imaging Another of Barnett’s examples showed a cheese sample where fat, protein, water and salt content are measured and presented in a map of those parameters, using just one image. “That’s all using fairly traditional line scanning cameras,” said Barnett, “but what I would like to see is a video camera where each pixel has a full spectrum behind it. You [wouldn’t] need to have a line looking at a conveyor belt. I can have a video camera that looks around a room and [says], ‘this is plastic’. That’s where we’re going with this technology now.”
Multispectral miniaturisation “The technology is moving towards miniaturisation as well,” said Barnett, “and this is more about multispectral. [Because,] once you know where the important parts are in your spectrum, you can just build cameras which have a few filters.” There are two different technologies – both “built in the semiconductor fab” – that allow manufacturers to build thousands of cameras at the same time and therefore drive down the cost of production. i
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39
