While people manually search for hogweed in the fields, the neural network is already doing it from orbit. The Yandex School of Data Analysis has launched a service that finds the dangerous plant directly on satellite images — quickly, accurately, and without human intervention.
The project was created in collaboration with the Yandex Center for Social Technology and the StopBorshchevik movement. Anyone can use it: from volunteers to farmers.
The difference in speed is enormous. What used to take hours to mark up, AI now does on average 50 times faster.
And there is something to look for. Sosnowsky's hogweed is a real invader: it grows like a virus, displacing all living things around it and can leave burns with simple contact.
The service has already shown results in real conditions. With its help, foci of infection with a total area of more than 400 hectares were found in 17 regions. The neural network has fully explored Moscow and the region.
And these are not just numbers. Volunteers travel to the coordinates, dig up plants and cut off inflorescences. For example, one of the large foci was quickly eliminated in the Plescheevo Lake National Park.
Everything works as simply as possible:
- if the map is already marked up — open it and look at the hotspots;
- if not — upload a satellite image (GeoTIFF), the algorithm itself finds problem areas.
Inside — Yandex computer vision technologies. The model was trained on 10 thousand images, checked for accuracy via IoU and additionally validated together with experts.
Further — more. In 2026, the service wants to scale up to 100 thousand square meters, including the Tver and Yaroslavl regions. In parallel, they will begin to create full-fledged contamination maps for reserves and protected areas.
And it won't stop there. The plans include teaching the system to find other invasive plants as well, and then collect a single threat map for the entire country.