AI-Powered Drones Could Tell Human Teams Where to Find Marine Debris
The National Centers for Coastal Ocean Science (NCCOS) at NOAA (National Oceanic and Atmospheric Administration), Oregon State University and their partners are developing a drone-based machine learning system to detect and identify marine debris along coastlines.
In December 2021, researchers used beaches near Corpus Christi, Texas, to evaluate the system's devices and refine detection methods. The research project began in June 2020 and will run until the end of May 2022.
Marine debris is a global problem that threatens the environment, shipping safety, coastal economies and possibly human health. Fast and accurate detection and identification of debris is key to cleanup and response actions that can prevent these impacts. Unmanned aircraft systems, or drones, offer this capability.
Polarized light (radiation whose waves oscillate in a single plane) reflected from artificial objects often differs from light reflected from natural objects such as vegetation, soil and rocks. Mounting a polarimetric camera on a drone can improve the detection of debris from the air. Researchers tested such a camera both on the ground and from a U.S. Coast Guard helicopter.
The helicopter flight allowed the team to simulate the altitude at which the drone system would fly and model what would happen if the drone used a polarimetric camera. The team then trained a machine learning computer program to find and classify debris in the collected images.
Once fully operational, data collected by the drone-based machine learning system will be used to create maps showing where marine debris is concentrated along coastlines, to guide rapid response and removal efforts.
Researchers will train NOAA's Marine Debris Program staff to use the new system and provide standard operating procedure guidance. The project is the result of collaboration between NCCOS, NOAA's Marine Debris Program, Oregon State University, ORBTL.AI and Genwest Systems.
As for detecting floating marine debris, researchers from the University of Barcelona developed an AI-based system last year that allows drones to do this as well.
With millions of tons of plastic washing into the ocean every day, scientists have no time to keep track of where it is, but they may gain a useful new tool. In February 2021, researchers from the University of Barcelona developed an algorithm that can detect and quantify marine debris using aerial photographs, and they hope it can work with drones to autonomously scan seas and assess damage.
Summing up the global problem of plastic pollution is a difficult task, as ever greater amounts enter the ocean every day and break into smaller fragments that are hard to track. The University of Barcelona team targeted these pieces floating on the surface, hoping to improve existing methods of tracking their distribution, which involve studying damage from aircraft and boats.
An interesting example of this is work carried out as part of the Ocean Cleanup project, in which research vessels went to the Great Pacific Garbage Patch and flew over it in aircraft equipped with sensors and imaging systems. They demonstrated a way to do this using infrared radiation to distinguish pieces of plastic circulating in the ocean from other ocean debris.
Instead, the University of Barcelona team turned to deep learning methods to analyze more than 3,800 aerial photographs of the Mediterranean Sea off the coast of Catalonia. By training the algorithm on these photographs and using neural networks to improve its accuracy over time, the team obtained an AI tool that could reliably detect and quantify plastic floating on the surface.
“A large number of images of the sea surface obtained by drones and aircraft during marine debris monitoring campaigns, as well as during experimental studies with known floating objects, allowed us to develop and test a new algorithm that achieves 80% accuracy in remote detection of floating marine macro-debris”
The tool can analyze images individually or sort them into different segments, counting the amount of debris in each section to estimate density. The system is an open-access web application available to professionals in the field, but the team plans to develop a version that can work with drones to fully automate the process.
“Automated aerial photography combined with analytical algorithms are more effective protocols for monitoring and studying this type of pollutant”
