DATE on the Edge: On-Site Oil Spill Detection and Thickness Estimation Using Drone-Based Radar Backscattering

Abstract

Oil spills are one of the most harmful maritime disasters known to man. It is important to promptly react to spill accidents for early detection and monitoring. Improving the effectiveness of monitoring techniques for oil spills helps mitigate their environmental damage to the ecosystem. In this work, we present a new edge-based approach for the accurate detection and thickness estimation (DATE) of thick oil slicks within the range of 1 to 10 mm. The DATE approach is based on U-net models that are designed to process multiple C- and X-band radar backscattering from drones under different ocean conditions. The models are trained on synthetically generated oil spill scenarios based on fluid-dynamic and Monte-Carlo simulations. Simulation results show a high probability of detection that exceeds 90% for all possible thicknesses even at low wind speeds, which is not the case when using state-of-the-art satellite-based synthetic aperture radar techniques whose performance significantly degrades at low wind speeds. For the thickness estimation, in terms of the IoU metric, our results significantly outperform state-of-the-art solutions by a factor of 2 under varying wind speeds between 2 and 8 m/s. Moreover, we implement the new models on a hardware compute platform to verify the feasibility of edge computing on platforms like drones for effective monitoring and tactical response. Implementation results show a maximum power consumption of 1 W that suits the limited power budget of a drone, and a short latency of less than 3 s to generate the DATE maps.

Keywords

• Edge computing
• oil spill
• radar backscattering
• slick detection
• thickness estimation
• u-net model