w FEATURE Machine Vision
Neural networks: more than just slicing swede
AI has arrived in the fresh produce industry and the results are astounding: significantly reduced waste and flawless product presentation, writes Paul Wilson, MD, Scorpion Vision
T
opping and tailing vegetables such as leeks, sprouts, carrots, parsnips and swedes is dirty work. Someone has to do
it, but, increasingly, no one wants to. This situation has been exacerbated in recent years as labour availability has plummeted, prompting the vegetable- processing industry to look to automation. Optical trimming systems guided by classic 3D machine vision technology have to date presented the best available option for automating this operation. However, they represent a compromise rather than a solution.
Classic 3D machine vision can only look for features that conform to a pattern or shape that is expected: the tip of a carrot or the stem plate on a leek, for example. However, vegetables don’t come in a fi xed size, shape or colour, and this inherent variability translates to compromised cutting performance. Taking leeks as an example, with an automation platform that relies on classic machine vision, just 60% of processed leeks will be cut at exactly the correct point. The rest will either be
16 September 2022 | Automation
cut roughly, at a poor angle or slightly too short or long. Some may be written off altogether or not cut at all. This level of waste is unacceptable from a commercial and an environmental standpoint.
Enter AI AI has introduced a new world of opportunity for revolutionising the performance and effi ciency of camera- driven cutting systems. In the leek- trimming example, it can be diffi cult to determine the stem plate when it is obscured by roots or debris. Not for a camera with AI, which confers the ability to look at and analyse each individual vegetable before making a decision on how to process it. The machine simply needs to be shown some examples of the stem plate in a variety of conditions and it will learn what to look for, enabling it to formulate its own conclusion about what it is seeing. AI does this by augmenting classic computer vision algorithms with models called neural networks. When a computer receives an image, machine-vision software compares that image data with a
neural network model – a process called “deep learning inference”. The net result is much more robust image processing. Stereo vision will enable the real-world dimensions of the product to be detected in the X, Y and Z axes, but overlaying AI enables the camera to recognise features that it wouldn’t normally. A machine- vision system based on AI can achieve repeatability of 99%, yielding a return on investment in a matter of months through waste reduction and yield improvements alone.
False economy Off -the-shelf cameras with built-in AI are widely available and attractive from a cost perspective but won’t match the levels of repeatability that we can guarantee with our bespoke systems; most will be lucky to achieve 80% reliability. That is because our experienced optical and automation engineers work together to build application-specifi c systems from the ground up. The starting point is the conception of a stereo camera array that can be used by the system’s software
automationmagazine.co.uk
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 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54 |
Page 55 |
Page 56 |
Page 57 |
Page 58 |
Page 59 |
Page 60 |
Page 61 |
Page 62 |
Page 63 |
Page 64
