Visual yield measurement of seed potatoes works

09-02-2022 | Crop solutions | Article

Harvesting of seed potatoes in the Netherlands (Noordoostpolder). - Photo: Henk Riswick

Automated site-specific visual yield measurement of seed potatoes works. This is evident after completion of a three-year trial in the Netherlands.

Seed potato grower Henk Geerligs in Anna Paulowna (the Netherlands) is excited about the possibilities of visual yield measurement on his potato harvester. The most recent version of the detection system was tested at his farm during harvest last year. Researchers at Wageningen University & Research are still busy processing all the data, but as far as Geerligs is concerned the results show that the prototype has a lot of potential.

Revenue model

Geerligs says he already sees a revenue model. “Let’s suppose such a system comes onto the market for, say, € 20,000. If I can optimise my cultivation with it, I will have earned that investment back in a few years.” He calculates: “One daughter tuber extra per meter of potato ridge, gives me € 250 per hectare. However, at the moment I don’t know what cultivation decisions I have to make to get those extra tubers. By doing simple practical trials with planting distance, fertilisation and seed treatment and then registering the yield during harvesting, I hope to find out,” says Geerligs.

However perfect and site-specific yield measurement may be, it is not possible to immediately attach a revenue model in the sense of costs and direct benefits.

‘Precise yield determination is essential’

Reliable site-specific yield measurement is an excellent tool for precision agriculture, for site-specific variation in crop growing. Based on common sense you may be able to reason out the effect of planting closer or wider, more or less fertiliser or irrigation, of varying the depth of tillage, but with yield measurement you know. Harvests trials also provide answers, but not as detailed as when looking at every square metre of the field.
Koen van Boheemen, scientist at Wageningen University, says precise yield determination is essential. “How can we as a university continue to recommend precision agriculture if we cannot demonstrate the effect of precise site-specific measures? If we want to demonstrate that precision agriculture really works, we cannot do so without objective and accurate yield measurements. This also applies to individual growers.“
If you mount such a device on your harvester, you won’t earn money immediately, Van Boheemen says. “But it does give an accurate indication of the effect of the variations you apply to your crop. The ultimate benefit then depends on which varieties and classes of seed potatoes you grow.“
For the trading company that buys the seed potatoes, it can also be interesting to know what he can expect from a certain grower even before sorting.

Looking at instead of weighing

At seed potato farms in Anna Paulowna and Wieringerwerf, research into visual yield measurement of potato yields ran in 2019, 2020 and 2021. “Visual” means acutally looking at how many potatoes are being harvested by the machine, instead of weighing them. And this is then done site-specifically, so that ultimately it is known exactly how many potatoes have grown at which location in the field. The latter is important to know which of Geerligs’ cultivation decisions produce extra tubers and which do not.

_{Text continues underneath image}

Koen van Boheemen, precision farming expert of Wageningen University & Research: "If you mount a device like this onto your harvester, you don't immediately start making more money. But it does give a precise indication of the effect of the variations you apply to your crop." - Photo: Jan Willem Schouten

Crop optimisation

The recent completion of the trials are certainly a step forward when it comes to cultivation optimisation. The objective of the project was to develop a system for very accurate determination of the net yield, tare and size distribution of harvested potatoes, more specifically seed potatoes. By now, there is a well-functioning prototype. Now, in the first months of 2022, the research project group is going to look at what steps to take next. Koen van Boheemen: “Maybe there will be a follow-up project. In any case, we want it to be put into practice and not remain on the shelf. It is not self-evident that a machine builder can immediately get to work with commercial kits. What we do know is that the visual yield determination on the DeWulf harvester from Geerligs works reliably. But that is just one machine. We need to know how it performs on other harvesters before we move towards commercialisation.”

The fact that visual yield measurement appears to be a promising technique for perfecting the cultivation of seed potatoes does not really come as a surprise, by the way. Van Boheemen already concluded after the first field experiments this technology is possible and has potential. Now, after harvest 2021, he is even more convinced. “We have now run the data from Henk Geerligs and we are tremendously close to the truth.”

Also for other crops

Visual yield measurement is currently being developed for seed potatoes, but in principle it is also interesting for measuring the results of (variation) trials with cultivation measures in potatoes for French fries, onions, beets or chicory. With visual site-specific yield measurement it is possible to determine the result of variation in size and planting distance in a broader way than with manual trial planting alone. French fries plants are very keen on a predictable flow of the most uniform potato tubers possible. This makes yield measurement a starting point for yield improvement.
Against this background, it would have been nice if more potatoes had been harvested by the contractor, as this could have reduced the costs of yield measurement. Incidentally, contract workers in the Netherlands report that sugar beet growers are willing to pay less than 10 euros per hectare, so to speak, to get an accurate yield map of a field.
Visual yield measurement is currently being developed for seed potatoes, but in principle it is also interesting for measuring the results of (variation) trials with cultivation measures in potatoes for French fries, onions, beets or chicory. With visual site-specific yield measurement it is possible to determine the result of variation in size and planting distance in a broader way than with manual trial planting alone. French fries plants are very keen on a predictable flow of the most uniform potato tubers possible. This makes yield measurement a starting point for yield improvement.
Against this background, it would have been nice if more potatoes had been harvested by the contractor, as this could have reduced the costs of yield measurement. Incidentally, contract workers in the Netherlands report that sugar beet growers are willing to pay less than 10 euros per hectare, so to speak, to get an accurate yield map of a field.

Each tuber is recognized

The most important component in this project is the 3D camera hanging above the reading belt of the potato harvester, which recognises each tuber individually and can reliably measure its size. In this way, data is collected regarding the net yield, the number of tubers and size of the harvested potatoes. With this data, growers not only receive information about the harvested potatoes in general, but via the GPS on the harvester also about where exactly they were grown. This creates a yield map of the field with a grid of 1.5 x 1.5 meters for a two-row harvester, or 3 x 3 meters for a four-row harvester.

Compared to weighing with load cells, visual yield measurement has advantages. For instance, the camera distinguishes potatoes from tare, which can expecialliy be of great advantage of clay soils. Moreover, with visual yield measurement you get a clear picture of the grading. This is particularly important for seed potatoes, since the exact number of tubers in the various sizes is more important than the absolute yield in kilograms. Precision farming technologies for seed potatoes are primarily aimed at getting as many kilos as possible in the more expensive seed potato sizes.

Yield measurement with weighing cells on the harvester does give an immediate indication of how many tons of potatoes go over the harvester. In situations where the soil tare percentage is fairly constant – on sandy soils it is virtually non-existent – site-specific weighing gives a good idea of the yield variation in a field. With visual yield measurement, the kilogram yield must be calculated afterwards. Although this could also be done in real time.

Camera above picking belt

Cameras on the harvester. Potatoes have to lie flat and evenly distributed on the conveyor belt, so the cameras can see all of them individually. – Photo: Misset

In the trial with visual yield measurement, one or two – depending on the width of the belt – cameras are mounted above the reading belt of a bunker harvester. These take pictures of all potatoes that pass by and link the images to GPS coordinates. A picture is taken of each filled belt. So a new picture is taken every time the belt moves by one camera frame. Depending on the speed of the conveyor belt, the number of photos varies from five per second with a fast moving belt to one or two per second if it is moving slowly.
After harvest, the images are analysed and the findings can be used to improve the next crop. The effect of variation in cultivation measures such as tillage, planting distance and fertilisation on grading thus becomes visible.
Analysing the images directly on the harvester itself is also possible in principle, but then every harvester would have to be equipped with a heavy and therefore expensive computer in addition to cameras. Moreover, a potato harvester is a dangerous environment for such a computer. Collecting the images and processing them later “in the cloud” is a lot cheaper.
Behind the camera on the harvester there must be a laptop with a large storage capacity linked to it if three high-resolution images are made every second. You can easily collect 300 gigabytes of images per day. GPS coordinates are linked to the images. Later in the process, an image of the sorting and tare in areas of 1.5 x 1.5 meters for a two-row harvester and 3 x 3 meters for a four-row harvester is created. The analysis gives a spot-by-spot picture of the size sorting, in which in principle common seed sizes are used. This is adjustable by the user. And there is also always a category smaller than 28, and usually it is 28-30 or 28-32 and then 30/32-35. Processing the images afterwards on a powerful computer requires no further manpower.
Much attention was paid in the final phase to improving the detection system. “The less interference, the better,” explains Koen van Boheemen of Wageningen University & Research. Interference can be light from outside. “Are you working at dusk or in bright daylight? During the research we tried to exclude differences in light by using artificial light under a hood. For this type of system to work properly, development in practice is necessary. Only then can you be sure it will work.“
Other improvements made to improve accuracy, according to Van Boheemen, include wet and dry harvesting conditions and color differences in the skin of potatoes. “The detection system has to learn to read all this, and that requires the necessary programming work. Before the algorithm can do its job properly, the system must be fed with enormous amounts of information.”