14:48, 21 January 2026

Russian Scientists Plan to Upgrade AI System for Real-Time Marine Litter Monitoring

The technology is also expected to be adapted for use on autonomous monitoring platforms.

Researchers from the Moscow Institute of Physics and Technology and the Shirshov Institute of Oceanology have developed an artificial intelligence–based system that can automatically detect floating marine debris and other objects from onboard a vessel, even in Arctic conditions. The system is designed to enable large-scale monitoring of pollution in the world’s oceans, the university’s press service told IT-Russia.

Frodo May Not Survive the Winter

Plastic pollution and other forms of marine debris are among the most serious threats to ocean ecosystems, alongside climate change. The Arctic region is of particular concern to scientists, as microplastics have already been found in the bodies of marine animals. Late last year, a large-scale rescue effort was launched in Kamchatka, a region bordering the Arctic, to save an orca calf named Frodo that had become trapped in a plastic ring. After two months, the operation was suspended, and experts warned that Frodo was unlikely to survive the winter.

Plastic has also been detected in seabed sediments. As for traditional methods of spotting debris on the ocean surface, such as visual observation, their effectiveness is limited. How many people would it take to visually monitor vast stretches of open water?

The researchers say they have found a solution. The system is built on two machine-learning approaches: image classification using contrastive learning and direct object detection. Both methods were tested on a unique dataset collected by the scientists during an Arctic research expedition in the fall of 2023.

“We processed more than 500,000 photographs of the sea surface taken in the Barents and Kara seas. The main challenge was the difficult shooting conditions – sea foam, ship motion, and extensive sun glare. All of this makes it harder to detect small objects on the water surface and at shallow depths. The system can identify four types of objects: marine debris, birds, glare on the water, and droplets on the camera lens. The development is particularly relevant for the Arctic, where pollution by non-degradable anthropogenic waste poses a growing threat to fragile ecosystems,” said Mikhail Krinitsky, one of the study’s authors and head of the university’s Laboratory of Machine Learning in Earth Sciences.

Good at Spotting Birds, Less So at Finding Trash

The most effective approach for detecting marine debris was based on contrastive learning using ResNet50+MoCo combined with a CatBoost classifier. The widely used YOLO algorithm performed worse in this task, although it proved more effective at detecting birds.

“The lower performance of YOLO may be linked to the fact that marine debris often consists of small objects that are hard to distinguish against waves. In addition, fortunately, trash is still relatively rare. A limited number of positive examples is a classic challenge for machine-learning models. Our approach, which involves pre-selecting image fragments, allowed us to better handle this feature of statistical learning,” added co-author Olga Belousova, a junior researcher at the same laboratory.

Looking ahead, the team plans to further refine the algorithms to enable real-time operation. The system is also expected to be adapted for deployment on autonomous monitoring platforms.

Earlier, IT-Russia reported that school students from the city of Rybinsk in Russia’s Yaroslavl region had invented an aquatic robot capable of collecting trash from the surface of the water.

Territory management and ecology