Automated Approach for Extraction of Oil Spill from SAR Imagery

Oil spill pollution is a major environmental concern since it has most dangerous and hazardous effects on marine environment. Periodic monitoring by detecting oil spills along with its movement, helps in efficient clean-up and recovery operations. Over the past few years, Synthetic Aperture Radar (SAR) based remote sensing has received considerable attention for monitoring and detecting oil spill due to its unique capabilities to provide wide-area observation in all weather conditions. However, the interpretation of marine SAR imagery is often ambiguous, since it is difficult to separate oil spill from look-alike features. The objective behind our study was to extract probable oil spill candidates automatically from SAR imageries containing oil spill incidences, where new methods based on over-segmentation and amalgamate approach is used for this purpose. The methodology is all about over segmenting the entire image based on its statistics and amalgamating relevant segments at later point of time to represent actual dark features as probable oil spill candidates. Under the dependency on SAR imageries alone, the approach does not take care of the separation of look-alike features which can be addressed subsequently by the consideration of associated synchronous external data resources such as optical data, wind and ocean parameters (Zhao et al. in Opt Express 22(11):13755–13772, 2014; Espedal and Wahl in Int J Remote Sens 20(1):49–65, 1999). The approach is carried out on a set of RISAT-1 imagery containing oil spill incidences and the extracted oil spill areas are in well agreement with the visually interpreted output with kappa coefficient greater than 0.70 and overall classification accuracy greater than 80%.