If you remove sunflowers too early, there are high losses during threshing and drying costs. If you are late, the seeds will crumble, birds will fly, fungi will develop. The golden mean is humidity of 25-30% for desiccant treatment and 10-12% for direct combining. But how do you know this moment when thousands of hectares are at stake?

The technology, for which a patent has already been obtained, allows to determine with 98% accuracy when the field is ready for desiccation (treatment of crops with special preparations (desiccants) that accelerate the drying and maturation of plants) and harvesting, simply by scanning the crops from a drone.

Experiments with sunflower are being conducted at the research laboratory for spectral phenotyping of agricultural plants, established in the Southern Federal University in July 2025.

At the moment, ten people work here, and more than half of them are young scientists: students and postgraduates of SFedU. The interdisciplinary approach has become a key feature of the team: not only biologists, but also IT specialists, mathematicians, and experts in neural networks have gathered here. Scientists write software for analyzing spectral data themselves using machine learning and deep learning algorithms.

"We can install software on the customer's equipment so that he can analyze everything himself, having unmanned vehicles on staff. For small farms, this can be an online service where the farmer himself flies through the field, downloads data and receives information about his field. It can also be a service for service companies that provide services to agriculture and have their own drones, expanding the range of their services. Another option is to sell a non—exclusive license or integrate our technologies as modules into existing precision farming platforms. There are many options, and we are working on them all," explains Pavel Dmitriev, head of the laboratory.

In practice, two main approaches are used today, and both have serious limitations. The first is visual: it is believed that a sunflower is ripe when the basket on the back side turns from yellow to brown, and the petals wither. But this is just a guideline: the moisture content of seeds in this phase can range from 10 to 18%, and for desiccation you need to get into the corridor of 25-30%. You can't tell such numbers by eye. The second is laboratory: specialists go out into the field, selectively mow down test batches of plants and take samples to the laboratory, where they are dried in special cabinets. The procedure has to be repeated several times per season, wasting man-hours, fuel and losing part of the harvest on experimental plots.

Drones are no longer exotic in Russian agriculture. Companies are actively using agricultural drones to monitor fields. However, most of the existing solutions work with vegetation indexes, such as NDVI, which perfectly show the general condition of plants, stress, drought, and diseases, but do not allow you to determine the moisture content of seeds by leaves or the general appearance of crops. Existing precision farming services provide, in fact, additional information about the field, but they cannot measure specific parameters: humidity, damage by phytopathogens, and the state of trace elements.

SFedU scientists have proposed their own solution: to look not at the seeds that are hidden in the basket, but at the back of the inflorescence. During maturation, its color and reflectivity naturally change along with the moisture content of the seeds. The experiment was conducted in the fields of the STEPPE agricultural holding in the Salsky district of the Rostov region. The researchers collected sunflower baskets of the Jinn M variety at different stages — from seed filling to full ripening. In the laboratory, using a hyperspectral camera operating in the 450-950 nm range, the spectra of the back side were taken, and then the actual moisture content of the seeds was measured on an analyzer.

To turn the spectra into a working tool, we used machine learning, a random forest algorithm. The model was trained on 70% of the samples and tested on the remaining 30%. The result is impressive: the accuracy of determining humidity on the back of the basket has reached 98% with an error of only 3-4%, which is quite sufficient for making agronomic decisions.

"We have proved that it is necessary to look at the back of the basket. The seeds are hidden, but the chemical processes in the tissues of the inflorescence are directly related to their maturation. Chlorophyll is destroyed, carotenoids change the ratio — and all this is visible to the spectrometer. The pigment-sensitive indexes helped us: CCI, Booch, Datt3, TCARI and others. They even picked up on the changes that the human eye cannot distinguish. The problem is that the agronomist determines the humidity accurately in relation to individual plants, but there is a big problem with a representative sample in order to correctly assess the situation throughout the field. Our technologies are designed to replace existing tools — not just to provide additional information, but to actually measure humidity in relation to the entire thousand plants in the field," says Pavel Dmitriev.

It is important that the method works after desiccation. When a plant is treated with drying preparations, chlorophyll is destroyed dramatically, and simple indexes like NDVI "go crazy." But the combination of nine indexes, selected by SFedU scientists, retains accuracy and allows you to control the dynamics of humidity right up to cleaning. Last year, scientists conducted bench tests that confirmed the possibility of transferring technology from the laboratory to the field. The measurement error of the new method was slightly more than one percent of the figures obtained by the classical method. The researchers are currently in the final stage of developing a prototype humidity detection system. This year, tests are planned on real fields with industrial partners.

Unlike point samples, the new technology allows you to get a map of ripeness for the entire field — not five to ten test points, but hundreds of thousands of measurements with reference to each square meter. This makes it possible to plan the order of harvesting: where the humidity is already twelve percent, then send the combines there, and where it is still eighteen— wait. Resource savings are also significant: there is no need to drive equipment across the field to mow trial lots, and there is no need to lose crops in experimental plots. To remove on time means to avoid shedding, fungal infestation and bird attacks.

Commercialization of the technology can take many different paths. For agricultural holdings, it is possible to install software directly from the customer: with unmanned vehicles on staff, agronomists will be able to fly over fields on their own, upload data to the program and receive an answer about humidity in a particular field and its distribution.

For small farms, an online service based on the SaaS model is suitable. When a farmer contacts a service company that provides agricultural services and has its own drones, they fly over his field, upload data to the cloud and receive information. For developers of precision farming platforms, it is possible to integrate technology in the form of modules into existing services.

The study was published in the international journal Seeds, and RF patent No. 2842590 was obtained for a method for remote determination of humidity. In the near future, scientists plan to expand to other crops: winter wheat, barley, chickpeas, and peas. The work was carried out with the support of the strategic academic leadership program "Priority 2030" (national project "Youth and Children"), which made it possible to update the instrument base and attract young staff.

"Thanks to the Priority 2030 program, we have the opportunity to upgrade our instrument base, obtain modern equipment for the development of competitive technologies related to hyperspectrometry, and the opportunity to attract additional personnel. It is impossible to simply go to the market, advertise and hire ready—made specialists to the laboratory - there are no such specialists. They can only appear inside the laboratory, and we involve undergraduates and postgraduates to work with them to grow specialists with the necessary competencies," adds the researcher.

Southern Federal University, being a participant of the strategic academic leadership program "Priority 2030" (national project "Youth and Children"), concentrates efforts on solving the tasks of scientific and technological development of the country. As part of this work, the university creates full-cycle production and technological chains based on the network architecture of interaction to respond to "big challenges". The key areas of development cover a number of critical and end-to-end technologies that underlie three key strategic technological projects of the university: "Technologies of soil bioengineering", "Technologies of multifunctional microelectronics and intelligent sensors for biohybrid and cyberphysical systems" and "Technologies for accelerated development and transfer of strategically important materials in micro and low-tonnage production".