Detection of garbage on the sea bed with artificial intelligence: DeepLitterAI

According to the Kyodo agency, researchers named the system they developed "DeepLitterAI".

The data set, created with approximately 12 thousand photographs taken by agency teams from the sea bed in the country between 1983 and 2025, includes non-garbage objects such as small garbage and rocks.

In order to prevent the artificial intelligence from detecting the wrong objects as garbage, the system was trained with various methods such as blurring.

The system, which can process data twice as fast as humans, will be used to simultaneously monitor plastic garbage on the seabed.

During testing with real images, the system can determine the type and amount of litter visible in an area as small as 5 to 10 percent of the image width. It can also successfully detect 80 percent of large pieces of garbage, such as plastic bottles and polythene bags.

Compared to visual inspection by experts, it was determined that the system's margin of error was around 10 percent and that the system completed the analysis in a few days, which people do in about a month.

"We can quickly identify areas where large amounts of waste are collected and use this information to implement measures," biological oceanographer Nakajima Ryota, one of the researchers, said in a statement. he said.

The research results were published in the journal "Environmental Pollution".