Detecting Floating Marine Macro Litter (FMML) Using Deep Learning Models

Floating plastic debris on water surfaces poses both immediate and long-term threats to the environment. Therefore, identifying and monitoring plastic pollution is crucial for understanding its location and scale. This article presents a framework for detecting and tracking floating plastic debris in inland waters using optical satellite image time series, leveraging the advantages of multitemporal Earth observation data. The detection process begins with a rule-based approach that analyzes variations in signal intensity, temporal patterns, spectral characteristics, and information fusion to identify potential plastic candidates. Once sensitive areas are detected, they can be continuously monitored, and the extent of plastic coverage at a subpixel level is estimated using spectral unmixing. The method requires only a specified time frame and area of interest as input parameters, eliminating the need for manually selecting specific images or outlining regions of interest. Several case studies demonstrate the successful application of this workflow, developed as a Google Earth Engine application, to identify highly affected areas in full Sentinel-2 scenes. These examples span different continents and environmental contexts, capturing floating plastic debris of varying types and dynamics.