Mapping marine litter with Unmanned Aerial Systems: A showcase comparison among manual image screening and machine learning techniques

Recent works have shown the feasibility of Unmanned Aerial Systems (UAS) for monitoring marine pollution. We provide a comparison among techniques to detect and map marine litter objects on an UAS-derived orthophoto of a sandy beach-dune system. Manual image screening technique allowed a detailed description of marine litter categories. Random forest classifier returned the best-automated detection rate (F-score 70%), while convolutional neural network performed slightly worse (F-score 60%) due to a higher number of false positive detections. We show that automatic methods allow faster and more frequent surveys, while still providing a reliable density map of the marine litter load. Image manual screening should be preferred when the characterization of marine litter type and material is required. Our analysis suggests that the use of UAS-derived orthophoto is appropriate to obtain a detailed geolocation of marine litter items, requires much less human effort and allows a wider area coverage.