Experts from the University of Barcelona (IRBio) and their colleagues have developed a new algorithm for the MARLIT open source web application that automatically estimates the amount of floating plastic in the sea using aerial photographs. Its work is based on deep learning methods. Measurement accuracy reaches 80%, according to a study published in the journal Environmental Pollution. The news appeared on the EurekAlert! Portal.

To develop the algorithm, the researchers analyzed more than 3,800 aerial photographs of the Mediterranean coast of Catalonia using artificial intelligence techniques.

Floating marine macro-debris is a threat to the conservation of marine ecosystems around the world. The largest density of floating debris is found in the great ocean gyres – systems of circular currents that spin and pull the debris with them. However, there is a lot of polluting waste both in coastal waters and in semi-enclosed seas such as the Mediterranean.

How much garbage has accumulated in the oceans is usually calculated using direct observations: from a boat, from an airplane. However, it becomes difficult to handle large amounts of data in “manual mode”. Therefore, there is an alternative – aerial photography combined with analytical algorithms. This – more automated – method also has its drawbacks. Factors such as waves, wind, cloud cover often hide floating debris from the probes.

Therefore, Spanish scientists decided to improve the automated method using artificial intelligence. “The large number of images of the sea surface obtained by unmanned aerial vehicles and aircraft during campaigns to monitor marine debris, as well as during experimental research with known floating objects, allowed us to develop and test a new algorithm that detects floating marine debris with an accuracy of 80 % at remote sites, ”says García-Garin, member of the Department of Evolutionary Biology, Ecology and Environmental Sciences at the University of Barcelona.

The new MARLIT algorithm allows you to analyze images individually, as well as divide them into several segments according to user instructions, determine the amount of floating debris in each defined area, and estimate the density of debris using image metadata (height, resolution). In the future, scientists plan to adapt the application to a remote sensor (such as a drone) to automate the remote sensing process.