SUPER-RESOLVING WASTE DUMPS FROM SPACE WITH DEEP LEARNING: ROMANIA REGION USING SENTINEL-2 AND SPOT6/7

An increase in the total quantity of waste produced on a global scale has severe repercussions for the ecosystem, including health risks for individuals and environmental damage. This increase is directly correlated with urban development in recent decades, making all member states of the European Union (EU) responsible for complying with waste management regulations. The detection of waste dumps represents the essential component of waste management, and satellite data provide the ability to monitor it. However, expensive, very high-resolution images are needed for proper identification. This paper proposes a super-resolution (SR) workflow to increase the readability of low-resolution but accessible satellite data (Sentinel-2). We assess the workflow for the specific use case of waste dump detection in Romania (including seven major cities). We also analyze several dataset pre-processing techniques and 18 popular Deep Learning (DL) models, providing valuable insight into super-resolution applications for waste management.