Automated floating debris monitoring using optical satellite imagery and artificial intelligence: Recent trends, challenges and opportunities

Abstract

Unwanted and harmful floating debris creates aesthetic, economic, social, and ecological harm. The optical satellites provide frequent global coverage across multiple spectral bands. Utilizing this abundant multi-banded optical satellite data for floating debris monitoring, many artificial intelligence-based approaches were proposed. These approaches face various challenges due to the multidimensional nature of the earth observation data visualized on a reduced scale. This work identifies various stages of AI deployment for floating debris identification, classification, segmentation, density estimation, and/or temporal study. The challenges during each stage along with some potential solutions applied in this field or elsewhere have been identified. Since AI approaches are data-driven, the limitation of labeled data with real-time diversity of shape, color, texture, size, and composition of floating debris placed against different backgrounds is most acute. The work proposes the utilization of some recent AI-based systems, like continuous learning, transfer learning, attention-based transformers, explainable AI, etc., to resolve these identified challenges. The work calls for further research into the application of pre-trained models, semi-supervised learning, and multi-modal data fusion for overcoming the labeled data deficiency. Additionally, harmful debris density estimation and factors leading to a change in the estimated density need further research.