Detecting and Classifying Marine Plastic Debris from high-resolution multispectral satellite data

Plastic debris in the global ocean is considered an essential issue with severe implications for human health and marine ecosystems. Remote sensing is a useful tool for detecting and identifying marine pollution; however, there are still few studies and benchmark datasets for developing monitoring solutions for marine plastic debris detection from high-resolution satellite data.

Here, we present an annotated plastic debris dataset from different geographical regions, seasons, and years, including annotations for sea surface features (e.g., foam), objects (e.g., ship) and floating macroalgae species such as Sargassum. Our dataset is based on high-resolution multispectral satellite observations collected mainly for the period 2014-2020 over the Gulf of Honduras (Caribbean Sea). Over this region, large plastic debris masses and Sargassum macroalgae blooms have been frequently reported, suggesting that it is an ideal region to examine satellite sensors' effectiveness in plastic debris identification, as well as monitoring along with sea surface circulation and meteorological data.

We also present a set of machine learning classification frameworks for marine debris detection on high-resolution satellite imagery, comparing qualitatively and quantitatively their overall performance. The new algorithms were validated against different regions that have been reported as major plastic polluted areas, as well as their performance was compared to well-established FAI and new promising FDI. This benchmark study can trigger more research and developement efforts towards the systematic detection and monitoring of marine plastic pollution.