Japanese scientists build AI tool to spot plastic waste on ocean floor

Researchers in Japan have developed a new artificial intelligence system designed to improve detection of plastic waste scattered across the seabed, according to local media reports published on Thursday.

The system, called DeepLitterAI, was developed by scientists at the Japan Agency for Marine Earth Science and Technology and can process visual data at roughly twice the speed achievable through manual human review, according to a Tokyo based news agency, which cited findings published in the journal Environmental Pollution. Researchers say the tool is intended for eventual use in real time monitoring of marine plastic pollution.

To train the system, researchers compiled a dataset of approximately 12,000 images drawn from decades of seabed footage collected by the agency dating back to 1983. The dataset included both actual litter objects and items that can be easily mistaken for waste during visual inspection, such as rocks and various marine organisms. The team trained the AI using a range of image variations, including blurred and inverted versions of the original photographs, in order to reduce the likelihood of false identifications.

Detecting plastic waste on the ocean floor has traditionally proven difficult, in part because a large share of the plastic that enters the ocean eventually sinks and settles among natural seabed features, making it hard to distinguish visually. Ryota Nakajima, a biological oceanographer involved in the research, said the new system allows the team to quickly pinpoint locations where waste has accumulated in large quantities, information that can then be used to guide cleanup and prevention efforts.

During testing using real seabed footage, the system proved capable of identifying both the type and volume of litter present, even in cases where individual objects made up as little as five to ten percent of the total image width. Researchers found that the AI successfully detected roughly 80 percent of major litter items, including plastic bottles and polythene bags, with an error margin of about 10 percent compared with assessments carried out by human experts through direct visual inspection.

The development comes as researchers around the world continue to search for more efficient methods of tracking ocean plastic pollution, a problem that scientists say has grown substantially in scale over the past several decades and now affects marine ecosystems at nearly every depth. Automated detection tools like DeepLitterAI could eventually help agencies and environmental organizations prioritize cleanup operations in areas where waste has concentrated most heavily, rather than relying solely on limited and time consuming manual surveys of the seabed.