Success story: using deep learning to grade seeds for Argirella Nervosa

Success story: using deep learning to grade seeds for Argirella Nervosa

21 - 12 - 2022

Deep learning is a high-precision data extraction and processing technique based on neural networks. Thanks to Rosepetal, seed producer Argirella Nervosa S.L. has applied this system for the classification of seeds using artificial vision.

Through deep learning we can detect defects in organic products and perform optical character recognition and product classification in accordance with size, colour, shape and any other kinds of characteristics.

It is a high-precision system that operates by using algorithms that simulate the functioning of the human brain. This is why we say it is based on neural networks.

Last year, Argirella Nervosa, a producer specialising in the organic seed sector (OEM), contacted us because it wanted to implement a seed grading system on its production line. 

What is Rosepetal?

Rosepetal is a suite of artificial vision software with supervised and semi-supervised learning that uses convolutional neural networks. 

It is highly reliable deep learning software created by the AIS group, suitable for any camera that the company already utilises. This is the system we successfully applied in the case of Argirella Nervosa for its seed grading. 

Although Rosepetal is a system normally used for organic products, due to their high degree of complexity (bread, biscuits, fruit, etc.), or manually assembled products (a tray with different components, an insert or a sticker), the possibilities offered by Rosepetal are endless.

What challenge did the customer pose?

Argirella Nervosa, an organic seed producer headed by Jairo Reig Boronat, needed a seed grading system that was able to operate using artificial vision. Only deep learning techniques could be used to guarantee success with this kind of product, given that it is so hard to inspect.

The aim of applying deep learning and artificial vision in this case was to automate the quality control without the need for a genetic operator to be constantly present. This person would thus only have to perform a final review of the data and then be able to devote his/her time to other more valuable and less mechanical tasks.

What was achieved by applying Rosepetal’s deep learning software?

A seed grading system was created using artificial vision. The seeds move along a belt one by one and, thanks to Rosepetal, they can be divided into four or five different groups so as to distinguish between them in terms of quality.

Therefore, automated quality control has been achieved by applying this software, with the inspection fully online. The system is fully integrated into production and can create a well-designed, error-free database. 

Before using Rosepetal, when visual and manual checks were performed, Argirella Nervosa estimated that there was a 20% margin of error and that there were errors in the database.

“The R&D put into deep learning has enabled us to take the seed selection process to a higher level. Now we can conduct our production more rapidly with all the inspection tasks on the line itself. This means that we can increase our business volume and improve the quality of our seeds.”

Jairo Reig Boronat

What did the process of applying Rosepetal involve?

As we have seen before, deep learning was the best solution for a highly complex kind of product such as seeds. 

However, before implementing this kind of deep learning system, we had to conduct a preliminary study on the implementation of an artificial vision classifier supervised by convolutional neural networks. This study is conducted with images taken by the customer or samples provided by a supplier. 

Based on the above, we can run tests and analyse the results obtained. We can thus predict the effectiveness of the actual application of a deep learning algorithm.

After this process, we can determine Rosepetal’s accuracy and sensitivity. In the case of Argirella Nervosa, we were able to confirm that this was the best solution.

Now that a year has gone by, what is your assessment?

Since its installation at Argirella Nervosa we have performed just one post-installation service on the software. Rosepetal has operated fully autonomously throughout the year with no incidents.