Japan research team develops AI system to detect plastic litter on seabed

A Japanese research team, comprising scientists from the Japan Agency for Marine-Earth Science and Technology, has devised an artificial intelligence system called DeepLitterAI. The system aims to more effectively identify marine plastic waste on the seabed. Published in the journal Environmental Pollution, the team's work presents a significant leap in plastic waste monitoring capabilities.

The AI, capable of processing data almost twice as swiftly as human analysts, processes footage of Japan's seabed collected since 1983. The dataset consists of roughly 12,000 images, featuring both litter and non-litter objects such as rocks and marine life. The team trained DeepLitterAI using various image manipulations, like blurring and inversion, to minimize false detections.

Ryota Nakajima, a biological oceanographer involved in the project, explained, "Our system can swiftly identify areas with substantial litter accumulation, enabling us to implement countermeasures more effectively." In tests using actual footage, the AI successfully identified the size and quantity of litter, including items as small as 5 to 10 percent of the image width.

It detected 80 percent of major litter items, such as plastic bottles and polythene bags, with a margin of error around 10 percent compared to expert visual inspection. The system can also complete an analysis that would typically take a human about a month in just a few days.

The significance of this new method lies in its ability to accurately detect both large and small pieces of litter. Since deep-sea footage is captured using wide-angle lenses, existing AI systems often struggle to recognize small litter items, often leaving them undetected. The new system has reportedly improved accuracy by 1.6 times compared to previous methods.