Identifying marine oil spills with thermal and hyperspectral UAV imagery: capabilities and challenges

Unpiloted Aerial Vehicle technologies have substantial potential in detecting marine oil spill incidents effective for emergency response and mitigation efforts. Offshore and onshore oil spill incidents remain under reported yet are frequently occurring and challenges remain around effective targeting of type-specific mitigation efforts. Here we demonstrate an approach of classifying imagery acquired from thermal and hyperspectral VNIR sensors mounted on UAV platforms of simulated oil spill incidents implemented in an outdoor environment under daylight conditions. Each dataset was classified using supervised pixel-based classification algorithms including Support Vector Machine, Random Trees and K-Nearest Neighbor. Six types of oil were successfully detected in seawater and in soil with SVM achieving between 74.69-76.09% overall accuracy, while RT achieving 67.77- 73.33 %, and KNN scoring 56.77-62.66%. This paper demonstrates that classification results of thermal infrared imagery had higher overall accuracy than hyperspectral imagery in detecting type specific oil which provide key insights about an oil spill incident necessary for mitigation efforts and recovery.