Detection and discrimination between oil spills and look-alike phenomena through neural networks

The proposed method is very promising in detectingdark formations and discriminating oil spills from look-alikes as it detects with an overall accuracy of 94% the dark formations and discriminate correctly 89% of examined cases.

Abstract

Synthetic Aperture Radar (SAR) images are extensively used for dark formation detection in the marine environment, as their\nrecording is independent of clouds and weather. Dark formations can be caused by man made actions (e.g. oil spill discharging) or\nnatural ocean phenomena (e.g. natural slicks, wind front areas). Radar backscatter values for oil spills are very similar to\nbackscatter values for very calm sea areas and other ocean phenomena because they damp the capillary and short gravity sea\nwaves.\nThe ability of neural networks to detect dark formations in high resolution SAR images and to discriminate oil spills from lookalike\nphenomena simultaneously was examined. Two different neural networks are used; one to detect dark formations and the\nsecond one to perform a classification to oil spills or look-alikes. The proposed method is very promising in detecting dark\nformations and discriminating oil spills from look-alikes as it detects with an overall accuracy of 94% the dark formations and\ndiscriminate correctly 89% of examined cases.