Offline-method spot spraying during the second herbicide application on sugar beet using the HORSCH Leeb PT
Future crop-protection regulations demand a sharp cut in the use of plant protection products. Spot spraying may be one possible answer. Theo Leeb and Josef Stangl explain where HORSCH stands on this, which approaches are already developed, and why all of this matters so much.
Back at Agritechnica 2019, several companies took part presenting startups with camera-based plant-recognition and spot-spraying solutions. There was considerable euphoria at the start. After all, systems like this are meant to enable precise application of crop-protection products, which is especially relevant and necessary given EU rules on reducing plant-protection product use. But how precisely and reliably does a spot-spraying system actually function? To find an answer to that question and offer precise solutions, HORSCH ran several trials.
Solutions that raised a lot of hope
 |
“At first glance, the solutions were very appealing, interesting, and market-ready. That’s exactly why we negotiated with three firms at the time, to see what stage things were actually at overall, and to check whether cooperation was possible. However, one of the companies leaned more toward fertilizer application, so those solutions weren’t especially interesting to us,” says Theo Leeb. |
Another startup focused on “green-on-green” and “green-on-black” spraying. In “green-on-black” technology, green corresponds to plant recognition regardless of whether it’s the crop or a weed, and black corresponds to soil. The “green” zone is the area where spraying will take place. In “green-on-green” technology, a camera system distinguishes between plant species – crop or weed – and the machine carries out targeted spot spraying accordingly.
“The ‘green-on-green’ topic is interesting to us in any case, particularly for the European market. We thought about which type of treatment this system might be useful for. The first thing that came to mind was thistle in wheat crops,” says Mr. Leeb. “Treatments like this are especially relevant when thistle spread is driven by crop rotation. Thistle is easy to fight in wheat – it grows in patches and is clearly visible,” explains Josef Stangl. Thistle can be considered a classic example of applying “green-on-green” spot-spraying technology.
An example application
“To test the reliability of recognizing and treating thistle in wheat crops during a trial, we tested one startup. Its system had a specific algorithm developed and trained just for this case.” To test it, large trial plots with varying degrees of weed pressure were laid out on HORSCH’s fields.
Theo Leeb’s realization came quickly. Although the system worked, its results were far from satisfactory. “Overall, spot spraying does function, but it has certain limitations. On average, the system recognized only 60% of the thistle, leaving the remaining 40% undetected. Of course, the question arises of how to interpret this result and whether this level of effectiveness is sufficient. Our conclusion: no, it isn’t sufficient. Results didn’t improve even after we further ‘trained’ the model with additional photos and information — treatment effectiveness stayed at the same 60%. So, for practical use, this system still isn’t mature enough.
Field of application
Mr. Leeb considers it important to determine where spot spraying will be most rational: on which crops, for which type of treatment, and during which period. It’s also important to take technical capabilities and limitations into account. “In my talks, I focus on visually determining the relationship between weed pressure and the smallest area that’s technically feasible to spray. The smallest area for a machine with a standard nozzle spacing of every 50 cm is 50 × 50 cm. If the distance between weeds (patches) is less than 50 cm, the sprayer’s nozzles won’t switch off. This means spot spraying only makes sense when weed patches in the crop occur rarely – at a distance of more than one meter apart.”
Let’s take corn as an example of a row crop and look at weed distribution. Depending on weather conditions, the distance between weeds is mostly in the centimeter range. A distance of several meters occurs only occasionally. Another problem complicating things overall is that weeds never stop germinating. “Today you treat a visible surface weed with a foliar herbicide, and next week a new one appears. To destroy weeds and manage late weed pressure, you need soil-applied herbicides. Foliar herbicides require more passes to achieve the same effect. In the end, a significantly larger area will end up being treated.”
These results point to the conclusion that, under these conditions, spot spraying has only very limited application. In row crops, applying soil herbicides provides “baseline protection.” Later, innovative technology can be used to precisely eliminate individual weeds. This approach, in Mr. Leeb’s view, can also be transferred to other row crops. “It’s harder with cereals, especially when it comes to grass weeds. It’s too difficult for an automatic recognition system to tell them apart from cultivated cereals. Recognizing meadow foxtail in wheat crops is almost impossible. It seems more realistic to me to recognize broadleaf weeds in cereal crops, since these plants differ noticeably in appearance. On the other hand, recognizing grass weeds in wheat in autumn or spring is currently an unrealistic task.” Spot spraying can only be implemented if weeds are clearly distinguishable from the crop. So it makes sense to apply it when weed pressure occurs only sporadically or in patches.
A spot-spraying system
In practice, there are cases where spot spraying delivers economic and environmental benefits. So HORSCH set itself the goal of developing its own spot-spraying system. “First we figured out which technologies it would be worthwhile for, and again came across the ‘green-on-green’ and ‘green-on-black’ topics.
A thistle patch in a sunflower field (5 × 5 m)
Various firms have already been applying “green-on-black” technology for quite some time, including in dry regions under zero-tillage, primarily in the US and Australia. That is, mostly where zero or minimum tillage is used after harvest. “Here, due to drought, weeds emerge weakly and sparsely. So fields are spot-treated with glyphosate. From a technological standpoint this is fairly simple, since it’s really just about color-processing a photo. The system checks what color a pixel is – green or brown (black). If it’s green, the nozzle opens and sprays the field. This can be achieved without using artificial intelligence. So this is our very first goal, since it forms the foundation for further technology.”
Artificial intelligence and camera systems
In parallel, HORSCH is working on artificial intelligence capable of recognizing plants. To train AI models, above all, a significant volume of relevant photo material is needed. The foundations the AI needs, and the appearance model, were developed in cooperation with the German Research Center for Artificial Intelligence (DFKI). Numerous photos labeling sugar beet and weeds were included in the AI model. “There are two approaches. In the first, individual weed plants are recognized; in the second, ‘only’ the crop is recognized. At first we focused on recognizing crop plants using AI. To achieve the most effective recognition quality possible, a large volume of photo material is needed. For further field treatment, these photos must contain images of plants at every important growth stage, under the widest possible range of lighting angles. For high recognition effectiveness, the system needs to be trained constantly. “Sugar beet was our test crop, which we used to train the system on. Anything recognized as beet wasn’t to be sprayed; everything else was identified as a weed and treated.”
Two methods
Besides the difference between the “green-on-green” and “green-on-black” approaches, spot spraying also differs by two methods of implementation. The online method involves a fixed camera mounted on the sprayer. Recognition, evaluation, and spraying all happen simultaneously during the working pass.
The offline method involves photographing the field before spraying using a drone. The photos are uploaded, evaluated, and used to create a georeferenced map with precisely marked spray zones.
These maps are then loaded onto the machine’s terminal, where they’re processed by the sprayer’s software. The time interval between the photo shoot and the treatment shouldn’t be significant, Mr. Stangl believes. Despite a good RTK signal, managing the processing of mapping data still remains something of a challenge.
 |
A significant advantage of this system is that the amount of working solution to be mixed is known in advance. Besides that, this option is comparatively cheap and helps get started with spot-spraying technology. And if we compare both methods, we’ll see that, thanks to its simplicity of use, the online method will become more widespread, Josef Stangl suggests. |
Where things stand today
“Spot spraying will never reach 100% weed-treatment effectiveness” – this is the unanimous view of both Josef Stangl and Theo Leeb. Some plants won’t be recognized by the camera, others fall outside the recognition model. So spot spraying’s effectiveness is always somewhat lower than that of full-coverage treatment. “There’s still a long road ahead. To get started with this technology, row crops are the best fit first of all – sugar beet, corn, soybean, sunflower, and possibly rapeseed. It’s easier for the technology to recognize the pattern in which these crops are arranged in the field, and so it more easily identifies the crop plants themselves. That’s why it also identifies weeds better. Especially if they grow in patches, like root-sprouting weeds, they can be located fairly precisely in row-crop fields. To reduce the use of chemical products, especially herbicides, a firm decision needs to be made to stop full-coverage treatment,” explains Josef Stangl.
Completing spot spraying of sugar beet using the HORSCH Leeb PT
Our goal is to release, in the foreseeable future, an effective “green-on-black” system for no-till regions. Beyond that, HORSCH is continuing to work on “green-on-green” plant recognition in row crops, in order to achieve crop-protection savings compared with full-coverage treatment. After all, in practice, farmers will be required to cut plant-protection product use by 50% over the next seven years. That fact alone could serve as a driver for developing technology in this field.
ABA “Astra” Agro-Building Alliance is one of the largest suppliers and service operators of agricultural machinery in Ukraine. The company’s core activities are the sale of new and used agricultural machinery, warranty service, and the sale of spare parts.
“Astra” offers its clients a wide range of machinery for practically every type of agricultural work: Fendt, Valtra, Manitou, Berthoud and Horsch, Framest, Bogballe, Valley, Trimble, Oros. The company supplies customers with the widest range of genuine spare parts to support all models of agricultural machinery from world-renowned manufacturers. Servicing agricultural machinery holds a key place in the company’s operations, covering nearly the entire territory of Ukraine and ensuring a fast response to farmers’ requests.
“Astra” has been operating on Ukraine’s agricultural market for 25 years, providing agricultural producers with quality machinery, innovative agricultural technologies, and a consistently high level of service
