Discriminating marine oil spill pollution by using ocean water surface texture

Marine oil spill can cause great damage to marine and costal ecosystems, so Synthetic Aperture Radar (SAR) is used to detect and survey marine oil spill. Yet SAR has limitations since low wind speed, seaweed and bad sea condition can also cause “look-alikes”. This paper proposes an approach of taking a closer look to those “look-alikes” and identify whether it is a real oil spill by analyzing the ocean surface images from aerial visible remote sensing. First, surface patches are binarized and features including spectral coherence, directionality fineness and strip transect are extracted; Second, oil spill models and clean water model are established; Finally, suspicious patches are checked by comparing each of its features to those from models to determine whether it is an oil spill patch. The efficiency of the algorithm is demonstrated by a preliminary experiment with 274 training patches and 185 testing patches.