Blogs

Deep learning optical grading systems - Ellips | Next Level Grading

Written by Mitchel Bakker | Sep 13, 2023 8:51:19 AM

 

“We are moving towards 100 percent detection accuracy, so that eventually sorting can be done without staff,” says Roland Scheffer of Ellips, which focuses on the development and optimization of grading and sorting technology for the fresh produce market. Together with colleagues Jody Bakker, Mitchel Bakker and Adriaan Vet, Roland indicates that Ellips has recently taken the next step in that process through the introduction of their new deep learning module, TrueAI, which is part of the new software platform, TrueSort 2.

Taking accuracy and visibility of specific defects to the next level

With TrueAI, Ellips ensures that so-called last-mile defects are detected. The world’s existing technology has difficulty identifying specific last-mile defects. These defects can occur in so many different ways, that it is extremely difficult to model and calculate all these combinations. The new software can detect these type defects through the use of algorithms that have been developed by leveraging vast amounts of relevant data in combination with the expertise of operators and quality control personnel.

Advanced grading systems: enabled by massive amounts of data

This development was not a simple job, Mitchel explains. “The challenge is in optimizing the system with large amounts of data; the more data that enters the system, the better it can detect a particular defect. This has been made possible because we have a large global customer base and have collected data during different seasons and conditions. This allows us to take the accuracy of detecting the tricky defects to the next level.”


Difficult to detect

An example of such a defect is the crack that can occur near the stem of  an apple. “These cracks can occur in all sorts of sizes and colors, depending on whether they have just been torn open or have been open for some time. In addition, the stem is always in the way and creates shadow, and there are only a few moments when you can look precisely into the stem area with the camera to see if there is a defect there,” explains Mitchel about the complexity of the defect. “Thus, it is very difficult to sort it out. With our TrueAI techniques, it is possible. Previously this was something that required the operator to establish complex parameters. However, since the defect can occur in many different forms, it is extremely difficult to manually optimize the settings.”

Software and operators together

Although the application of Artificial Intelligence (AI) in the sorting process increases reliability and reduces the need for employees, it does not mean that the role of the operator is eliminated. “The operator has control over the quality that is sorted out, however many of the parameter settings previously done by the operator are now handled by the system,” says Mitchel. “You want the best of both worlds: an operator who has knowledge of the product and how to target the right defects combined with the powerful new tools built into our TrueSort2 grading technology,” adds Jody. “Based on customer preferences, market demand, competition, etc., there will still be situations where the final establishment of quality settings is made by the operator.”

The new generation of grading systems

“This is the future, our technology is taking fresh produce grading and sorting to the next level,” says Mitchel. Our TrueAI deep learning software is currently applied to apples and dates, and development is underway for expansion to other commodities including onions, cherries and blueberries. In addition, Ellips has spent the last few years preparing to upgrade its customer’s existing grading systems with the TrueAI module.


Clear dashboards

At the same time we have been enhancing our grading capabilities through the incorporation of TrueAI, our TrueSort software platform has been given a makeover. This effort has improved the user experience while making it possible for our customers to more easily incorporate data from the sorting process into dashboards or ERP systems. “In each batch, you can then see the percentage of defective fruit, for example,” says Jody. “But also the percentage of waste and the breakout of the various and more quality aspects associated with their grading efforts. So, more measurement columns and classes also appear on the screen which does not benefit the clarity and ease of use. “With TrueSort2, we have ensured that the relevant data is grouped in columns in a more logical and clear form.

 

Sort directly to requirements

From this data, the user can make a selection if necessary and define a product by grouping classes. This also provides our customers with the ability to sort directly according to their customer’s requirements. “Pack2Spec ensures that the supermarket specification can be met at the exit. As our customers are able to sort more and more accurately, they are able to deliver exactly what the market expects,” says Roland. “That means optimizing the customer’s yield and profitability while also helping to prevent food waste.” One of our TrueSort2 users is US-based apple grower Star Ranch Growers. Mitchel says that they utilize a 12-lane Ellips/Elisam apple sorting line with 60 product exits, which makes breaking down different grades important. “Our system has provided them with the flexibility to address all of those packaging and quality requirements through the use of our TrueSort2 software platform in an organized manner.” Click here for the full news item on Starr Ranch