AI Trained in Russia to Detect Invasive Hogweed From Satellite Images

Yandex Data Analysis School, together with experts from the Center for Technology for Society and the StopBorshchevik movement, has launched a free AI-powered service that detects invasive hogweed in satellite images. Image annotation is up to 50 times faster than manual processing.

Sosnovsky’s hogweed spreads rapidly, displacing native plants and posing risks to human health. To detect infestations, users simply upload satellite images in GeoTIFF format. A computer vision algorithm analyzes the image and highlights areas where the plant is present. The model was trained in Yandex Cloud on a dataset of 10,000 images with infestation sites, with accuracy verified by StopBorshchevik experts.

Find and Eliminate

Volunteers tested the service in 17 regions of European Russia, identifying hogweed across 421 hectares. Moscow and the surrounding region have been fully mapped. Based on the findings, plants are removed and flower clusters are cut. Special attention is given to protected areas. For example, the service helped detect and eliminate a large infestation in the Pleshcheyevo Ozero National Park in the Yaroslavl region.

Volunteers will continue expanding coverage and updating the dataset. By summer, the service will cover 100,000 square kilometers in the Tver and Yaroslavl regions. It will also be used to prepare infestation maps for federally protected natural areas. In the future, a nationwide hogweed map is planned.

Yandex also plans to train the model to detect other invasive plants, including Echinocystis lobata and Canadian and giant goldenrod.