Monitoring Coastal Marine Debris Using High-Resolution Satellite Image Time Series

Abstract

Marine debris is a serious global issue as it threatens marine life as well as the environment. Debris accumulation in coastal areas is highly influenced by geographical location and weather phenomena. Therefore, its prediction is challenging. To monitor marine debris, we acquired a time series of high-resolution satellite images of the westernmost island of Japan, where several polluted beaches are located. At a biweekly frequency, 23 images were acquired in approximately one year. First, we extracted the area of interest for each beach using an unsupervised clustering method. Subsequently, to estimate the debris density, we calculated the entropy from the semantic segmentation model. The results were cross-checked with the drone images captured on corresponding dates. Finally, we predicted the evolution of debris density over time and location.