Coastal Marine Debris Detection and Density Mapping With Very High Resolution Satellite Imagery

Abstract

Marine debris is a serious problem for marine ecosystems and related coastal activities. We carry out a study using in-situ debris clean-up data (collected by a local Japanese company) together with high spatial resolution satellite images to determine how well the satellite images can be used to estimate the amount and type of debris deposited on the beaches of the island in southern Japan. We use machine learning techniques to analyze the satellite images and find that Shannon's entropy computed from World-View 2 and 3 imagery from Maxar Corporation yields a useful detection and mapping of the coastal debris when compared with the in-situ clean-up data. We also assign a debris concentration to each satellite image pixel to visualize the distribution of the debris. The algorithm linking the satellite images to the ground truth clean-up data can now be used in areas, where no ground truth data are available.

Keywords

• Anomaly detection
• machine learning
• marine debris
• satellite imagery analysis
• semantic segmentation
• very high resolution satellite imagery