Aligned with World Environment Day 2025’s focus on ending plastic pollution, this open-sourced solution can help environmental agencies and volunteers accelerate solid waste removal, including plastics, in ecologically sensitive zones worldwide.
India’s plastic pollution crisis
India is the world’s largest plastic polluter, with 9.3 million tonnes of plastic waste annually, of which 126,513 metric tonnes reach oceans via rivers. This kind of pollution poisons ecosystems, disrupts food chains, and threatens marine biodiversity and human health. For instance, The Gulf of Mannar, home to over 4,200 species, faces coral reef threats as plastic waste suffocates and blocks sunlight, endangering over 8% of reefs. The Sundarbans, the world’s largest mangrove forest, receives 40,000 tonnes of plastic annually, disrupting breeding grounds and reducing local fish, shrimp, and crab catches by up to 15%, directly harming local livelihoods.
A significant part of the plastic waste accumulates along India’s 11,098-kilometer coastline, with reports revealing that 90% of waste found on the country’s beaches is plastic, including bottles, caps, and polystyrene products. Managing this waste is often challenging due to the remoteness of polluted areas and difficulties in estimating the number of volunteers and equipment needed for cleanup.
With machine learning automating waste detection and analysis, the neural network developed by Yandex and FEFU researchers now streamlines pollution assessment, offering a faster, cost-effective alternative to outdated methods — a critical step in combating the marine crisis globally.
Proven impact and opportunities for global adoption
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 an optimal number of volunteers and determining the pieces of equipment needed for the cleanup.
Further project development in 2025 includes deployments across Far Eastern and Arctic national parks, where challenging terrain complicates waste management.
Addressing the pressing issue of pollution, this solution can be further developed and implemented by local volunteer teams and government agencies in India and other countries with coastal areas, riverbanks, and similar environments, enabling more effective solid waste monitoring and cleanup. Additionally, having an open codebase, it can be customized to detect new types of waste, monitor endangered species, and support other environmental efforts.
How the AI solution works
The AI solution development leveraged computer vision, specifically semantic image segmentation, to automate solid waste detection. This method divides images into pixel groups, assigning each to a specific waste type: fishing nets, iron, rubber, large pieces of plastic, concrete, and wood, achieving over 80% accuracy.
The neural network then maps waste locations, estimates volume and weight, and calculates the required workforce and equipment (for instance, dump trucks, all-terrain vehicles). This data-driven approach optimizes logistics, reducing cleanup time and costs.
The neural network can be integrated with various mapping tools, such as the open-source QGIS.
The neural network codebase is fully open-sourced and available on GitHub. Environmental agencies and volunteer organizations around the world can use the model for free and modify it for their own pollution management tasks.
About Yandex
Yandex is a global technology company that builds intelligent products and services powered by machine learning. The company aims to help consumers and businesses better navigate the online and offline world. Since 1997, Yandex has been delivering world-class, locally relevant search and information services and has also developed market-leading on-demand transportation services, navigation products, and other mobile applications for millions of consumers across the globe.
.jpeg)
.jpg)
.jpg)
.jpg.jpeg)
.jpeg)
