Assessment of Marine Debris on Hard-to-Reach Places Using Unmanned Aerial Vehicles and Segmentation Models Based on a Deep Learning Approach

Abstract

It is difficult to assess the characteristics of marine debris, especially on hard-to-reach places such as uninhabited islands, rocky coasts, and seashore cliffs. In this study, to overcome the difficulties, we developed a method for marine debris assessment using a segmentation model and images obtained by UAVs. The method was tested and verified on an uninhabited island in Korea with a rocky coast and a seashore cliff. Most of the debris was stacked on beaches with low slopes and/or concave shapes. The number of debris items on the whole coast estimated by the mapping was 1295, which was considered to be the actual number of coastal debris items. However, the number of coastal debris items estimated by conventional monitoring method-based statistical estimation was 6741 (±1960.0), which was severely overestimated compared with the mapping method. The segmentation model shows a relatively high F1-score of ~0.74 when estimating a covered area of ~177.4 m2. The developed method could provide reliable estimates of the class of debris density and the covered area, which is crucial information for coastal pollution assessment and management on hard-to-reach places in Korea.