Abstract

Plastic pollution represents a growing threat to global ecosystems and human well-being, creating an urgent need for advanced methods that support accurate monitoring and mitigation efforts. Recent technological developments have transformed environmental assessment practices, especially through the use of aerial systems combined with artificial intelligence. The integration of unmanned aerial vehicles with deep learning techniques offers considerable advantages by enabling the coverage of large areas while providing rapid and precise analysis of collected data. The You Only Look Once family of deep learning models has gained prominence for real-time object detection, tracking, and counting, making it suitable for large-scale environmental applications. This study evaluates multiple versions of the You Only Look Once model to determine their effectiveness in quantifying plastic litter on coastal beaches using aerial imagery. The performance of versions 8, 9, 10, and 11 is assessed in terms of detection, tracking, and counting accuracy using a custom dataset consisting of short aerial video sequences. All evaluated versions demonstrated high counting performance, with accuracy values exceeding 98%. Version 11 achieved the strongest results overall, including the shortest inference times. These outcomes highlight the potential of advanced deep learning–based detection systems to enhance automated environmental monitoring and support more efficient management of plastic pollution on coastal environments.