Recent studies (OSIP REACT) showed promising results from applying Spectral Unmixing and Artificial Intelligence (AI) on fused satellite multi/hyperspectral data for detecting floating accumulation of plastic offshore. No significant results were achieved detecting plastics on beaches or near shorelines. Further studies are needed. The present idea explores AI potential for remotely sensing debris in dumpsites close to waterways and floating matter accumulations by focusing on two different nested modules, exploiting Sentinel-2 (S2) and PlanetScope (PS).
1st module is devoted to producing improved S2 data through spatio-temporal fusion techniques exploiting PS spatio-temporal fine resolutions and S2 spectral resolution.
2nd module is devoted to detection of debris in dumpsites close to waterways and floating matter accumulations on the new, improved data. 2 custom-based CNN architectures are investigated: ResNet, U-Net
Several open-source marine-debris databases provide the training/evaluation datasets (e.g., Marine LitterWatch, OceanScan, EMODnet Marine Litter Database, Litterbase).
