AI Spots Hidden Sea Litter at Record Speed

A Japanese research team has unveiled an innovative artificial intelligence system designed to enhance the detection of plastic waste accumulating on the seabed. Developed by scientists at the Japan Agency for Marine-Earth Science and Technology, the system, named DeepLitterAI, processes visual data at nearly double the speed of human analysis.

This breakthrough technology was trained using a substantial dataset of approximately 12,000 images captured from Japan's seabed since 1983. The system was specifically engineered to differentiate between plastic litter and natural elements like rocks and marine creatures, significantly reducing misidentifications.

Tests using actual deep-sea footage demonstrated DeepLitterAI's remarkable ability to detect litter items as small as 5 to 10 percent of the image width, achieving an 80 percent success rate for major items like plastic bottles and bags. The AI's margin of error is approximately 10 percent, a substantial improvement over existing methods.

Ryota Nakajima, a biological oceanographer involved in the project, highlighted the system's utility in quickly pinpointing areas with high waste accumulation. This capability is vital for implementing effective countermeasures against marine pollution, addressing a challenge that has long been difficult to accurately assess due to the nature of deep-sea environments.