INTEGRATED MONITORING SYSTEM FOR BEACH LITTER PREPAREDNESS AND RESPONSE

Abstract

The detection of the coastal and Marine Litter (ML) using UAS data in combination with machine learning methods is an important step towards an automated process of detecting and mapping ML concentrations. The Visual Geometry Group-19 (VGG19) CNN architecture is used to classify UAS image tiles in two classes; litter and no litter. Testing the geographical transferability of our method to an unseen dataset, we found that the VVG19 CNN obtained an overall accuracy of 83.67 % and an F-score of 81.63%. The produced ML density maps can be used as a decision-making support tool. The integration and visualization of the marine litter and density information facilitate decision-making by all related to the problem stakeholders and decision-makers.