SpotSpraying – What is the current status?
The future framework conditions in crop care provide for a significant reduction. A possible solution: SpotSpraying. Theo Leeb and Josef Stangl tell us what the current status is at HORSCH, which methods there are and why it is mandatory.
Already at the Agritechnica 2019, there were some companies and start-ups that dealt with the topic of camera-based individual plant detection and SpotSpraying and presented respective solutions. The initial euphoria was considerable. For these systems were to allow for a precise application of crop care agents – a requirement especially with regard to the stipulations of the EU for reducing crop care agents. But how well and how precisely does SpotSpraying acutally work? To find answers to these questions and to be able to present good, pinpoint solutions, HORSCH, too, did some intensive research.
Promising solutions
“At first glance, the solutions seemed very attractive, interesting and even very market-ready. This is why at that time we got in contact with three companies to see what the status of the project was and if there was a possibility for co-operation. The focus of one company, however, was rather on fertilisation. So this solution was not that interesting for us“, Theo Leeb comments.
But another start-up from France also dealt with the topic “green in green” and “green in brown”. With “green in brown” green corresponds to the plant, regardless if cultivated plant or weed, and brown corresponds to the arable soil. All areas detected as green are sprayed. With the “green in green“ principle, individual plants, weed or cultivated plants are detected by the system and sprayed in a targeted way.
“The topic “green in green” is definitely interesting for us, especially for the European market. We debated which would be the obvious application options for this system. And we soon ended up with the topic of removing thistles in wheat by spraying”, Theo Leeb says. This is particularly interesting if thistles are a rotation-related problem. “In wheat, thistles can be fought easily, usually come in nests and are easier to detect”, Josef Stangl explains. Thus, thistles would be a typical case for a “green in green” spot application.
Application
“To find out how reliably the thistles are detected and removed from the wheat by spraying, we tested the system of a start-up which had developed and trained an algorithm for this application case.” HORSCH set up and examined large test plots with different thistle populations.
For Theo Leeb, disillusionment came quickly. Though it worked, the results were far from satisfying. “In principle, spotting works. But there are limits. On average, only 60% of the thistles were detected, 40% remained untouched. Given this fact you wonder how to assess this result and if this degree of efficiency is already sufficient. Our decision was: it is not. And it did not improve after we continued to train the model with graphic material and information – the hit rate remained at approx. 60%.” This is still far from market-ready.
Field of application
According to Theo Leeb it is important to consider in which crops, for which kind of application and at what time SpotSpraying really makes sense. What is possible from a technical point of view resp. which restrictions there are has also to be taken into account. “When I hold a speech I like to point out the ratio between the smallest technically possible spot size and the weed population. The smallest spot size approx. amounts to 50 x 50 cm with a standard nozzle size of 50 cm. If the distance between the weeds is within the range of 50 cm, the sprayer will not switch off. This means: spotting only makes sense if the weed population is relatively thin – namely with distances of more than one metre.”
If you take the row crop maize as an example and take a look at the weed distribution, the spacings between the weeds usually are in the range of centimetres, depending on the weather. Spacings of several metres between weeds are rare. Another problem that complicates the situation is that weeds do not stop emerging. “If I treat the visible weeds with a leaf-active herbicide via SpotSpraying today, new weeds will grow next week.To regulate the weed regrowth and a late weed infestation, you need soil herbicides. If you only rely on leaf-active herbicides, it requires several passes to achieve the same result. In the end, you have treated a larger area.”
The conclusion was that SpotSpraying under these conditions only makes very limited sense. In row crops, you can achieve a “basic protection” by applying soil herbicides on a large area. Later, weed that emerges individually can be removed by means of spotting systems. According to Theo Leeb, this approach can also be transferred to other row crops. “In cereals, however, it becomes more difficult especially if you also take care of grass weeds. For the detection systems, it is extremely hard to tell them from cereals. The detection of black grass in cereals is almost impossible. What I could imagine is the detection of broad-leaved weeds in cereals. For in this case, the plants can be distinguished clearly. But to tell grass from wheat in autumn or in spring – this is not realistic today.” SpotSpraying is only feasible if the weeds can easily be distinguished from the crop, and it only makes sense if weed infestation occurs sporadically or in nests.
SpotSpraying system
There are application cases where spot application is positive from an economic and an ecological point of view. This is why HORSCH’s objective is to develop its own SpotSpraying system. “First of all, we checked which technology was behind it – and again we came back to the topics “green in green” and “green in brown”.”
The topic “green in brown” has been on the market for quite some time from different suppliers and is mainly used in the US and Australia in dry regions resp. direct seed regions. Mainly in regions where no or only little tillage is carried out after the harvest. “Due to the drought, there is only little weed emergence of individual plants which then are spotted with glyphosate. From a technological point of view this is rather simple as it is a mere processing of images. The system checks if a pixel is green or brown. It it is green, the nozzle opens and sprays. This can be carried out without artificial intelligence, and it is one of our first objectives as it is the basis for the rest of the process.”
AI and camera systems
In parallel, HORSCH works on an artificial intelligence which can detect plants. To train the AI models, appropriate and comprehensive graphic material is required. The basics that an AI requires and what a model looks like were prepared together with the DFKI (Deutsches Institut für künstliche Intelligenz = German Institute for Artificial Intelligence). A multitude of labelled images where sugar beet is distinguished from weed were included in the AI model. “There are two approaches. In one, the individual weeds are detected, in the other one, “only” the cultivated crop is detected.
Initially, we focus on the detection of the cultivated plant by the AI. To achieve the highest possible detection safety requires a lot of images. The graphic material has to include all relevant growth stages at different lighting conditions for the future time of use. A good detection performance requires continuous retraining. “Sugar beet was the test crop we trained the system for. For the beet is not treated and everything else that is green is weed and is sprayed.”
Two methods
Beside the “green in green” and “green in brown” differentiation, SpotSpraying can also be divided into two different methods. For the online method the camera is permanently installed on the crop care sprayer. Detection, analysis and application are carried out during the pass in one work process.
In the offline method, the image is recorded beforehand by means of a drone. In the next step, the image data is analysed and a geo-referenced map with the exact weed locations is prepared.
These maps are then loaded to the terminal and processed by the sprayer as application spots. According to Josef Stangl, the time between drone flight and application should not be too long. Despite RTK, the map management still is a challenge. An important advantage of this system is that you know in advance how much spraying mixture you need. Moreover, this version is relatively low priced if you want to enter the sector of SpotSpraying. If you compare both systems, Josef Stangl speculates that the online method will establish itself due to the easier handling.
Current situation
“SpotSpraying will never hit 100 % of the weeds“, Theo Leeb and Josef Stangl agree. Some plants will not be detected by the camera, others will slip through the detection model. Thus, the degree of efficiency for SpotSpraying is somewhat less than for an all-over application. “There is still a long way to go. Row crops like sugar beet, maize, soya and sunflowers, perhaps also rape, are ideal for starting with this technology. In these crops, the technology can recognise a pattern and thus detect the cultivation plant more easily. Weed can be identified better. Especially weeds that come in nests like our root weeds can be located relatively reliably in row crops. To achieve a reduction especially in the sector of herbicides it is a must to say: I do no longer spray the whole field”, Josef Stangl explains.
The objective is to have a guided “green in brown” system for traditional direct seed region in the foreseeable future. Moreover, HORSCH continues to work in the sector of “green in green” for individual plant detection in row crops to make use of the saving potential compared to the field sprayer. The fact that in seven years an agent reduction of 50% has to be achieved, can be a motivator in this respect.