Litter detection and mapping from the combined use of multispectral UAV imagery and Deep Learning: A case study from Greece

Litter pollution has grown to be the most prominent threat to the coastal ecosystems, affecting both the environment and the local communities. An important step towards the mitigation of coastal pollution is the effective monitoring of the issue. The rapid evolution of Remote Sensing has offered many new techniques for the detection of beach litter, and Unmanned Aerial Vehicles (UAVs), especially, have proven to be invaluable tools. In this study, different approaches of beach litter detection are evaluated in order to determine which ones yield the most promising results. The data used were collected in the area of Palio Faliro, Greece and included RGB and Multi-spectral images. For the detection of the litter from the UAV images, two Deep Learning (DL) models were utilized, namely the Mask R-CNN and the YOLOv3. The accuracy of these two DL models in beach litter detection and also explore the potential challenges that may arise while trying to monitor the coastal environment with UAV methods. Our study findings suggest that the combined use of DL methods and UAV imagery can provide a cost-effective and scalable solution in litter detection and can assist relevant decision-making actions. Future work will focus on evaluating different DL methods under other experimental settings as well which will help towards assessing the wider applicability of the combined use of drone imagery and DL approaches in litter detection in coastal areas.

KEYWORDS: Remote Sensing, coastal little, UAVs, drones, deep learning, ACCELERATE project

Acknowledgements 

This study is financially supported by the ACCELERATE MSCA SE program of the European Union’s Horizon research and innovation program under grant agreement No. 101182930