Yandex and FEFU Open-Source Neural Network for Accelerated Coastal Waste Cleanup
TL;DR
Yandex's open-sourced neural network offers environmental agencies a competitive edge by enabling cleanup operations four times faster than traditional methods in remote areas.
The neural network utilizes semantic image segmentation to classify waste types with over 80% accuracy, optimizing cleanup logistics by calculating required resources and equipment.
This technology significantly reduces plastic pollution in ecologically sensitive zones, safeguarding marine life and improving the health of our planet for future generations.
Discover how Yandex's AI transforms coastal cleanup, turning the tide against plastic pollution with a tool that's as innovative as it is impactful.
Found this article helpful?
Share it with your network and spread the knowledge!
Yandex B2B Tech, Yandex School of Data Analysis, and Far Eastern Federal University (FEFU) have open-sourced a neural network designed to streamline coastal waste cleanup in hard-to-reach regions, addressing the critical global issue of marine pollution. This technology, successfully deployed in the remote areas of South Kamchatka Federal Nature Reserve and now being tested in the Arctic, aligns with World Environment Day 2025's focus on ending plastic pollution. The open-sourced solution enables environmental agencies and volunteers worldwide to accelerate solid waste removal, particularly plastics, in ecologically sensitive zones.
More than 11 million tons of plastic—approximately 85% of all marine waste—enter the world's oceans annually, primarily as food packaging and fishing gear. Without intervention, this amount could surge to 29 million tons by 2040. Marine plastic pollutants degrade rapidly into microplastics when exposed to sunlight and seawater, threatening wildlife that ingests them. Discarded fishing gear, known as "ghost nets," accounts for 60% of ocean debris and lethally traps marine species. While much waste remains in oceans, significant portions wash ashore in remote coastal areas, where volunteers often avoid due to inaccessible terrain and uncertain resource requirements.
Traditional mapping of polluted areas relies on labor-intensive manual geotagging, limiting scalability. The neural network automates waste detection and analysis using computer vision, specifically semantic image segmentation, which divides images into pixel groups and assigns each to a waste type—fishing nets, iron, rubber, large plastic pieces, concrete, and wood—with over 80% accuracy. It then maps waste locations, estimates volume and weight, and calculates the required workforce and equipment, such as dump trucks and all-terrain vehicles. This data-driven approach optimizes logistics, reducing cleanup time and costs. The neural network can be integrated with various mapping tools, including the open-source QGIS.
During expeditions in Kamchatka's nature reserves, the neural network revealed that 33–39% of coastal waste was plastic containers and packaging, while 27–29% derived from industrial fishing. By deploying the tool, volunteer teams cleared 5 tons of waste four times faster than traditional methods, mobilizing optimal volunteer numbers and determining necessary equipment. Further project development in 2025 includes deployments across Far Eastern and Arctic national parks, where challenging terrain complicates waste management. The solution can be implemented by local volunteer teams and government agencies in Indonesia and other countries with coastal areas, riverbanks, and similar environments, enabling more effective solid waste monitoring and cleanup.
The neural network codebase is fully open-sourced and available on GitHub, allowing environmental agencies and volunteer organizations worldwide to use and modify the model for free. This adaptability supports customization to detect new waste types, monitor endangered species, and bolster other environmental efforts, marking a significant advancement in combating the global marine crisis through technology-driven solutions.
Curated from News Direct
