UNSUPERVISED LEARNING APPLIED TO MINERAL OIL SPILL AND BIOGENIC OIL DISCRIMINATION IN RADARSAT-2 IMAGE

Accurate detection and characterization of offshore oil slicks are essential for decision making in prospection and environmental contexts. The similarity of oil spills from different sources makes their classification in SAR images challenging. In this work, we presented an oil spill clustering model based on RADARSAT-2 image. We developed a model able to distinguish between mineral and biogenic oil based on a state-of-art self-supervised deep clustering algorithm. The Silhouette score and Calinski-Harabasz index were used to define the optimal number of clusters.