As plastic pollution continues to pose a threat to our oceans, a team of researchers from Wageningen University and EPFL has developed a groundbreaking artificial intelligence (AI) model to significantly improve the accuracy of detecting floating plastics in satellite images. The study, recently published in iScience, showcases the AI-based detector's ability to operate effectively even when faced with challenges such as cloud cover or hazy weather conditions.
Plastic waste is a growing concern globally, with significant amounts ending up in rivers and lakes before making their way into the oceans, forming clusters with natural materials like driftwood and algae. The new AI model, designed to estimate the probability of marine debris in satellite images, holds promise for systematic removal efforts, potentially using ships to target plastic hotspots.
The research team leveraged freely available Sentinel-2 satellite images, which capture coastal areas worldwide every 2–5 days. The volume of data generated by these images requires automated analysis through AI models, specifically deep neural networks. Marc Rußwurm, Assistant Professor at Wageningen University, explained that the model was trained using examples provided by oceanographers and remote sensing specialists who visually identified thousands of instances of marine debris in satellite images from various global locations.
The AI-based marine debris detector, developed following data-centric AI principles, estimates the probability of marine debris for each pixel in Sentinel-2 images. The training method, coupled with a refinement algorithm, allows the model to better predict marine debris objects, even under challenging conditions like cloud cover and atmospheric haze.
Rußwurm emphasized the detector's accuracy in difficult scenarios, crucial for situations where plastic debris is often washed into open waters after rain and flood events. The study highlights the model's potential application in monitoring and tracking plastic debris in real-time, combining weekly Sentinel-2 images with daily PlanetScope acquisitions for continuous monitoring. The dual perspective provided by both satellite platforms also enables the observation of drift directions, offering valuable insights into the movement of debris influenced by wind and ocean currents.
With plastic pollution being a persistent global challenge, the development of advanced technologies like this AI model offers a promising step towards more effective monitoring and mitigation strategies for protecting our oceans.
