Neural Networks Take On Ocean Plastic Along Russia’s Coastlines
The “Chistyy Bereg” (Clean Shore) project has been running for several years. Over that time, volunteers have been training artificial intelligence to help remove waste from Russia’s coastlines. A neural network developed with Yandex has been refined to operate with high accuracy, and the tool is now fully ready for environmental use.
Each year, around 11 million tons of plastic enter the world’s oceans. By 2040, that figure could rise to 29 million tons. Russia has one of the longest coastlines in the world, stretching roughly 40,000 kilometers. Many sections, including remote areas, are heavily polluted with marine debris. Covering such vast territory with volunteer efforts alone is extremely difficult.
“The problem was that we could see the waste, but we didn’t understand its scale. How much is there? What volume? What weight?” said Roman Korchigin, project director at the Nature Defenders Foundation and head of the Chistyy Bereg initiative.
Artificial intelligence was brought in to address this gap. The neural network was first tested in the Kronotsky Nature Reserve in Kamchatka. More than 6,000 drone images were collected, which the system annotated and used to identify key sources of pollution. Plastic accounted for about 39% of the waste, while industrial fishing debris made up 29%. These insights helped volunteers organize cleanup efforts more effectively and, importantly, reduce the time required by a factor of four.
“We carried out cleanup operations and waste analysis, collected data, and then used it to train the neural network,” Korchigin explained.
This “digital environmentalist” does more than spot waste in images. It calculates precise coordinates, identifies composition, estimates weight, and even forecasts the scale of cleanup work. Since 2024, volunteers have cleared more than 50 kilometers of coastline in Kamchatka and Primorye. During expeditions in 2025, the system was trained on an additional 20,000 aerial images. With grant support from the Presidential Nature Foundation, the project has expanded geographically.
Making Seas and Shores Cleaner
The technology is now set to be deployed across ten protected natural areas, including the Arctic. Rough estimates suggest that around 1,200 tons of debris are currently floating in Arctic waters. Researchers from the Moscow Institute of Physics and Technology and the Shirshov Institute of Oceanology have developed an AI-based system capable of automatically detecting marine debris on the surface of Arctic waters. Installed on vessels, it analyzes live camera feeds in real time, identifying foreign objects among waves, glare, and spray. The system is designed to operate in harsh marine conditions and can detect even relatively small objects, which is critical for early-stage pollution monitoring.
Meanwhile, students at the Polytechnic Institute of Sevastopol State University have developed an unmanned catamaran for collecting waste at sea. Dubbed the “Marine Scorpion,” it can operate up to 500 meters from its operator, rotate 360 degrees, and gather debris into a dedicated net.
From Coast to Coast
In 2026, Chistyy Bereg will expand to include national parks such as Kurshskaya Kosa, Land of the Leopard, the Commander Islands, Arctic territories, and the Dagestan Nature Reserve. The neural network will be used to plan the deployment of volunteers and equipment more efficiently, estimating in advance how many people and machines are needed. That means faster, more targeted cleanup operations across vulnerable coastlines.
“It is crucial for scientists to study how waste moves and to survey plastic accumulation on the seabed, as this directly affects the future of the Arctic, the North, and the species that live there. Neural networks are key to enabling research in harsh Arctic conditions, where work is difficult and resources are limited and require careful planning,” said Artem Smolokurov, head of the Clean North – Clean Country movement.
The data annotation methods, cloud architecture, and computer vision models developed for the project could be applied by any country with an extensive coastline.
