Researcher pair drones, machine-learning against marine litter

A project in support of the National Oceanic and Atmospheric Administration (NOAA) is pairing sensor-equipped drones with machine-learning applications in an effort to automate the identification and mapping of marine waste and create effective methods for its collection and disposal. 

NOAA’s National Centers for Coastal Ocean Science (NCCOS) and Oregon State University are leading the drive to construct an integrated system to spot and identify large volumes of marine waste. Drones flying with polarimetric cameras capture images that are fed into a machine-learning computer program to identify, classify, and map the marine debris in collected images. The final purpose of the asset will be to assist the fast and effective clean-up of the floating or beached litter, which often injures and kills marine life, interferes with navigation safety, and poses a threat to human health. 

Human-made pollution in the seas and along coastlines is a massive and growing problem around the world. Each year billions of pounds of trash end up in oceans, including at least eight million tons of plastic – one of the worst and most enduring forms of waste. On average, 33,000 single-use plastic bottles are dumped into global waters each minute, compounding the millions more already polluting marine life and environments. 

The NCCOS program seeks to mitigate the effects of that waste to oceanic ecosystems by pairing imagery captured by drones with the machine-learning computer capabilities to analyze and map different kinds of marine trash. Using that, the campaign will try and create operationally viable procedures and workflows suitable for consistent implementation by NOAA’s Marine Debris Program.

Though still a work in progress, the project conducted tests last December, overflying beaches near Corpus Christi, Texas to evaluate UAV performance and refine detection methods involved. 

The trials equipped UAVs with polarimetric cameras, which may prove more effective in detecting marine waste from the air. Those sensors pick up the differences in polarized light reflecting from human-made objects like plastic and metals compared to vegetation, soil, rocks, and sand. The data of the non-natural debris filmed by drones is run through the machine-learning program, which over time trains itself to identify different kinds of marine liter and place it on maps with indications of volumes involved. 

It’s believed the process will be particularly effective in organizing fast removal of large masses of waste that come to shore after severe events like hurricanes and tsunamis. But it also seeks to establish procedures for NOAA and similar organizations around the world to locate and systematize methods for dealing with accumulated marine litter. 

To that end, the NCOOS project will also continue to test and compare new drone sensors and platforms; link them with machine-learning applications to determine where substantial accumulation exists and requires collection; overlay marine debris concentration maps with priority habitat areas; develop and trial operational procedures; and combine data on waste concentration and material type to inform removal strategies and priorities.

The NCOOS program is one of the more recent and farthest-reaching deployments of drones to battle marine waste. Others have included non-governmental organizations flying the craft to map the estimated 1.6 million kilometers of trash known as the Great Pacific Garbage Patch, and a Geneva-based non-profit group using UAVs to create a global map of marine plastic pollution.

Photo: Amy Uhrin, NOAA