Artificial intelligence
Experience unparalleled precision in your packhouse with deep learning for apples, blueberries, cherries, and citrus fruit
What do we mean when we say, "LUCAi™ knows the difference"?
Leaps in accuracy for even difficult-to-classify defects happen because LUCAi™ classifies by looking at the entire piece of fruit.
By inspecting the whole fruit and understanding the entire fruit in context, LUCAi™ knows the difference between two similar defects and applies the correct classification.
Traditional grading using machine learning works by looking at different parts of the fruit – a few pixels at a time – and making a prediction based on its local training set. It does not see the whole fruit simply looking locally at whether areas of pixels are good or bad, rot or not rot.
LUCAi™ is different from other types of machine learning: it’s a full deep learning classification model. In short, LUCAi™ models mimic the visual cortex of the human brain, which is the area of the cerebral cortex that makes sense of what your eyes “see”.
These characteristics makes it the ideal solution to scan and recognize an image and put together all the individual characteristics the optical system “sees” before making a smart prediction about the entire piece of fruit.
Then, LUCAi™ will assign:
With pixel-detail nuances and richer information data, LUCAi™ makes smarter decisions in areas that otherwise take additional work to get right (e.g., defects in the stem bowls of apples, near the stems of citrus, packman and stem/nose cracks on cherries, dehydration for blueberries).
LUCAi™ not only arrives in your packhouse with increased performance, but its capabilities allow for a simple set-up. Because TOMRA has taken the effort to pre-train LUCAi™ it makes smarter predictions immediately, it only requires simple instructions from the operator. Operators learn to use LUCAi™ in hours and remain in control of your final sort parameters.
LUCAi™ uses a multi-layer artificial neural network, which extracts and summarizes the key information in the image. Those layers give it the accuracy to identify any defects because it’s capable of handling contextual data, comparing it, and seeing it in the wider context (defect severity).
With the help of a TOMRA optical grading system, LUCAi™ uses extremely fine differentiation that can outperform even the most seasoned human grader.
Freshco, a leading New Zealand fruit and vegetable grower and exporter, shares the benefits they've experienced with the implementation of the Spectrim LUCAi™ upgrade package on their apple line. Spectrim with LUCAi™ includes new ‘plug-and-play’ Deep Learning models, pre-trained by vast datasets, that can precisely detect and classify even the trickiest apple defects, such as splits and punctures across multiple varieties.
"We deal with a lot of Royal Gala and Breeze, and the tend to have a lot of splits. This year, after a storm, the fruit seemed to have a lot more splits. AI made it really, really easy to control that. Production was not slowed down at all like in the past, which means through put stayed at a good place within the shed." - Robin Mudgeway
Prima Frutta are based in Linden, California and pack apples, cherries and walnuts with their TOMRA sorting and grading solutions. The family-owned company is passionate about delivering quality to their customers and achieve this through their investment in technology.
"We're in the process of teaching it powdery mildew defects, insect bites, freshwater cracks. These things were very difficult to pick up our traditional mapping systems. LUCAi™ has seemed to be able to get these types of specialized defects more effectively."
