Automatic discrimination of hydrocarbon spills from ASAR images using a support vector machine (SVM)

Oil spill detection over ocean areas using synthetic aperture radar (SAR) images is a complicated operation due to the presence of other phenomena with signatures similar to those of oil slicks, and hence, different automatic or semi- automatic detection systems have been proposed in order to distinguish the real oil spills. In this work it is proposed a classifier based on a supervised learning method named Support Vector Machine (SVM). The algorithm was developed using 26 ENVISAT ASAR images acquired during the Prestige oil spill at the end of 2002 on the north- west coast of Spain. These images show not only a great number of oil slicks but also a lot of false alarms (or look- alikes). With the aim of training and validating the classifier, it was necessary a priori categorization of the signatures using other data sources, including direct observations that allow us to verify several slicks as oil. Results show a high degree of accuracy (98.1%) and significance (93.5%) in the validation, and also promising generalization capabilities of the algorithm.