Battling the harmful Cercospora fungus with big data, light and artificial intelligence

Joint work in the field

The researchers from the Technical University of Clausthal and the Research Center of Jülich are testing two different approaches in the Bavarian field – without competing with each other. The opposite is actually the case. They are working toward a joint solution to a “global problem,” as Willer says in describing the Cercospora fungus. The project is being funded by the German Ministry of Agriculture.

On this particular day, the scientists have chosen exactly the right time to be standing among the test plants: This part of Germany is warm and humid between the end of July and the beginning of August, and the plants have become so big that their leaves touch. These are the optimum conditions for the fungus to spread.

“Each of us provides a piece of the puzzle”

During this initial project measurement work on an approximately one-hectare field, the researchers from both institutions harvested about two terabytes of data with their two pieces of test equipment. This was enhanced by weather data that are essential for plant growth: humidity, wind, wind direction, precipitation, air pressure and temperature. All the data is then evaluated and processed by another project partner – the Technical University of Dortmund. “With our measurements, each of us provides a part of the puzzle that will form the complete picture in the end,” Willer says.

Affected areas warm up differently

As part of work to develop an automatic evaluation process that would enable early recognition of Cercospora, the Technical University of TU Clausthal decided to use a special laser whose infrared beam is invisible to the human eye. In the process, the sugarbeet leaves are illuminated by the beam. “The infected leaf areas are warmed differently than the healthy ones at certain wavelengths in the mid-infrared range,” Willer says.

An infrared camera records the illuminated leaf of the plant, measures its temperature in the process and saves the data. Ultimately, the infection of the plant should be visible through the infrared camera – before the human eye can detect the infection.

Plan: a small, mobile system

The field test represents the first measurements conducted by the researchers from Clausthal outside the laboratory. In contrast to lab conditions, in Plattling they have to deal with sunlight. Willer says this presents a special challenge because solar radiation affects the measurement results depending on its intensity. The researchers initially measure the temperature of healthy and infected leaves without using the laser and then with it. The temperature difference is stored as a data set. This means one thing: “If the sun is shining directly on a leaf, the base temperature will already be higher and the difference following exposure to the beam may not be so meaningful,” Willer says.

The scientists are also working to streamline the use of the measuring devices on the field: Plans call for the equipment to be more mobile and compact when it is used on an upcoming project in Italy, Willer says.

Ultimately a tractor will tow the equipment across the field; it won't have to be pulled by hand. By that point, the researchers also want to have determined the optimal wavelength for their system.

Computer scientists train artificial intelligence

The data from the initial measurements conducted outside the laboratory, the weather data and the Cercospora model data are now in the hands of computer scientists at the Technical University of Dortmund. They are sifting out the relevant data and teaching computer software to spot infected plants.

But which data is relevant? And how does a computer learn to recognize it? In the case of the Clausthal measurements, for instance, the experts in Dortmund are tracking the wavelengths of the laser at which the temperature difference recorded on the infected plants is the most striking. This difference is communicated to the software as an indication of infection. This step is repeated, and in the process, the artificial intelligence algorithm is trained to determine on its own whether a plant is infected.

Jülich uses plant illumination

The Dortmund specialists have also received the data collected by the team at the Research Center of Jülich led by phenotyping expert Dr. Onno Muller. A Jülich system called LIFT (light induced fluorescence transient) illuminates the test plants with light at a particular wavelength just like the Clausthal system.

But the approach is different from the one used by scientists in Clausthal. The researchers in Jülich are focusing on photosynthesis in the leaves. Or to be more exact: They are working with chlorophyll fluorescence. Chlorophyll is the natural product of photosynthesis.

Does the fungus affect photosynthesis?

It works like this: Photosynthesis, or the conversion of the sun’s light energy and carbon dioxide into chemical energy in the form of glucose, begins when the light is absorbed by the green pigment called chlorophyll. In the process, part of the energy absorbed by the plant is released again and re-emitted. This “glow” is called chlorophyll fluorescence. The researchers in Jülich record it with the help of the LIFT sensor, Muller says.

His hypothesis: If a leaf is infected with Cercospora, the infection will have an impact on photosynthesis in this area. Because chlorophyll fluorescence is closely related to the efficiency of photosynthesis, the altered radiance could be a sign of fungus infection on the affected leaf areas.

The following analysis shows how the data gathered by scientists from Clausthal and Jülich can be optimally combined. If the temperature of the affected area of an infected plant rises to a certain degree value, and if it re-emits a certain amount of energy on exactly this area, these findings will serve as two indications of Cercospora infection on the same area of the leaf.

Later on, this information should be automatically identified and reliably analyzed by computers using algorithms known as machine learning or artificial intelligence. A condition of achieving the high planned recognition rate is that the system be trained with known images. In the DataPlant project, these will be images of infected and healthy leaves. The team of physicists and cultivators will note on each photo whether it shows healthy or infected leaves and thus create a huge reservoir of training data for artificial intelligence. And because huge amounts of data are involved, the team speaks of “big data.”

“This process is not limited to Cercospora,” project head Christoph Bauer says. “We think that this principle can be employed to automatically spot many different leaf infections at an early stage – and we are laying the foundation for this right now.”

The interplay of new types of sensors, conventional photo analysis and machine learning has not been confined to pure research at KWS for a long time now: These technologies are already being used in cultivation and research.

Developments like these will provide farmers with faster access to different varieties, and rapid assessments of the health of their crops will also be possible, for instance using drones.

The project partners:

The DataPlant project is being funded by the German Ministry of Food and Agriculture. Why? The key reason is that the project is being conducted by a consortium of leading institutions and companies in Germany that are involved in the complex area of “digitalization of agriculture.” In addition to KWS, the group consists of seed experts, physicists at the Technical University of TU Clausthal, phenotyping experts at the Research Center of Jülich and computer scientists at the Technical University of Dortmund who have joined KWS in the investigation of new data-analysis opportunities. The project also includes the companies Infratec and MG Optical Solutions for sensor and measurement technology as well as experts from BASF Digital Farming.