Using unmanned aerial vehicles (UAVs) and machine learning techniques for the assessment of Posidonia debris and marine (plastic) litter on coastal ecosystems

Due to the dynamics of the littoral zone and continuous inputs from tidal oscillations, riverine sources, anthropogenic activity, and inputs from extreme events, the use of unmanned aerial vehicles (UAVs) for assessing marine litter and Posidonia deposits on beaches has become a necessity. In recent years, Posidonia oceanica beach wracks (banquette) along Mediterranean coasts have increased and are often associated with marine litter items, mostly plastic. In our study, we attempted to list marine litter and categorize the items from pictures The results show an abundance of plastic debris at the Kerkennah site, where the average value of plastic deposits exceeds 100 items per 170 m of beach length. In the second step, we used UAVs for 3D mapping of Posidonia deposits and identification of marine litter classes. The orthomosaic was used for rapid Posidonia volume quantification on beaches. Machine learning was employed to assess MML (Marine-Macro Litter) abundance based on the orthophotos assembly. In the following step, we proceeded with ortho-mosaic creation and delimitation of the study area for the identification and analysis of macroplastics and other items using processed data. Our findings and image processing demonstrate that the K-nearest neighbor (KNN) approach yields excellent results compared to in situ quantification and assessment along the Kerkennah transect for four tests of macro plastic identification and quantification. The error of using UAVs was calculated with an average value of 6.3%, with bottle shape identification being the dominant result.