Onion packing centres updated about Eqrafts Eqrader


Last week, the onion sector was brought up to date about upgrades made to the Eqraft optical sorting machine in the past six months. This was done at Dutch onion trader, Hoza, who have the first model of the Eqrafts Eqrader.

This machine sorts onions by size, colour, form, weight, and internal as well as external quality. Feedback received at a similar event six months ago, was included in the advancements.


The machine at Hoza als gave a lot of insight into how an onion season runs. This input led to further developments of the machine. There have been various mechanical adjustments. The machine has been optimally fine-tuned so the captured camera shots are in sharp focus, regardless of their position. This is all important when it comes to accurate sorting.

Bas Pomstra, Head of R&D at Eqraft, shared which mechanical improvements had been made. He mentioned the conveyor belt, a so-called 'V-band', which places the onions in cups and guides them through the machine. "By changing the way this V-band is constructed, there is a more streamlined transition to the cups. An average of 75% to 90% of the cups are filled", says Jim Hoogzand of Hoza.

The machine has a tilting section, in order to check the onions on all sides. The problem with this was that the smaller onions could slip through the chains of cups, which caused mechanical failures. This problem was solved by making a minor adjustment to the machine.


The machine has two integrated cup washing systems. The cups transporting the onions get dirty after a while. The washing system automatically washed these cups. It is advised to wash the cups once a week, but this depends on the batch of onions that has passed through the machine. The new Tally version's interface has been adjusted. Users find it useful to have so much relevant information displayed on one screen. Real-time information of the sorting process is important. Setting are displayed in the background.

The Q eye camera now also features new software, and there are two additional cameras. This means there are now six cameras aimed at one track. This gives even better coverage of the onion, as a whole. The focal distance has been optimally fine-tuned. This means images of onions of any size can be captured in sharp focus, in order to properly judge the quality of the onion.


Besides the mechanical improvements, significant progress has been made in the development of the machine's neural network. The neural network receives data based on a classification model that has been developed. This classification model comprises characteristics used to determine the quality of the onions. The external quality characteristics are divided into 18 categories. Examples of these are: Good - a perfect, or acceptable, onion; Skinned - some cracks, or half or fully skinned; Stains - mild, or heavy stains. These are seven of the 18 external characteristics. You can see a complete overview of these characteristics in the photo.

Bas Pomstra asked those present if these 18 characteristics cover all possible variations. He also asked that onions from a particular batch be checked and the participants then place them in a particular category. This caused much discussion, making it clear that choosing the correct category is no easy task, and that opinions vary widely when it comes to choosing categories. The Eqrader was much quicker, sorting ten onions per second per track!


“The user can sort batches that meet certain quality characteristics needed for a particular customer or country.”

Bas Pomstra - Head R&D / Eqraft


The neural network has been 'taught' based on these categories. The user can sort batches that meet certain quality characteristics needed for a particular customer or country. The neural network is a system that requires further development in order to achieve optimal sorting. The software must 'learn' the different characteristics. Pomstra, therefore, called for the neural network to be further developed with help from the onion sector.

There are currently insufficient onions in each of the classification model's categories to supply the neural network with sufficient data. The goal is to collect red and yellow onions in all the categories, as quickly as possible, and to do further tests.


The first results look good. From a batch of 100 000 onions, 80 000 were used for training. These 80 000 onions are used to teach the machine how to distinguish between the various categories. The 20 000 remaining onions, which had not yet been sorted, were used in an 'exam'. The results were good, and came close to the minimum capacity of the traditional readings done on the onions. In the matrix, you see, for example, the reading for stones is at 100 in the diagonal line. This means all stones were removed from the batch.

After Rutger Keurhorst and Bas Pomstra's presentation, there was the opportunity to see the Eqrader in action. It is a very nice process, where the onion are placed one-by-one into the cups. Above the cups is a Modesta dusting extraction system, which blows onions skins and dust away. After the onion is checked, internally, it move through the sorter in order to be checked externally.


After this is done, the onion is turned and placed in a different cup, where the other side is also checked. The sorter now "knows" which onion this is, and categorises it. Crates, containing the desired types of onions, are filled from six different dispensers in the machines, which are fed by an output system. The dispensers can be adjusted as per the customers wishes, and sorting demands. If the customer wants perfect onions of two different sizes, then two dispensers are used.

The afternoon's conclusion was that the mechanical and electronic developments of the Eqrader are progressing well. Advancements have been made in the past six months. The development of the neural network, using the classification model, has led to the development of a reliable system which can index the onions' various quality characteristics.

Source: freshplaza.com / Author: Andries Gunter

Find out more about our solutions

Tijmen van de Poll