A neural network approach to oil spill detection using SAR data

Topouzelis, K; Karathanassi, V; Pavlakis, P; Rokos, D

dc.contributor.author	Topouzelis, K	en
dc.contributor.author	Karathanassi, V	en
dc.contributor.author	Pavlakis, P	en
dc.contributor.author	Rokos, D	en
dc.date.accessioned	2014-03-01T02:49:22Z
dc.date.available	2014-03-01T02:49:22Z
dc.date.issued	2003	en
dc.identifier.uri	https://dspace.lib.ntua.gr/xmlui/handle/123456789/34552
dc.relation.uri	http://www.scopus.com/inward/record.url?eid=2-s2.0-22344442574&partnerID=40&md5=408d2036035e5e1db92a6d9406beb8dc	en
dc.subject.other	Classification algorithms	en
dc.subject.other	Marine environment	en
dc.subject.other	Oil spill detection	en
dc.subject.other	Training algorithms	en
dc.subject.other	Algorithms	en
dc.subject.other	Backpropagation	en
dc.subject.other	Data reduction	en
dc.subject.other	Environmental engineering	en
dc.subject.other	Gravity waves	en
dc.subject.other	Image segmentation	en
dc.subject.other	Imaging systems	en
dc.subject.other	Multilayer neural networks	en
dc.subject.other	Synthetic aperture radar	en
dc.subject.other	Oil spills	en
dc.title	A neural network approach to oil spill detection using SAR data	en
heal.type	conferenceItem	en
heal.identifier.secondary	IAC-03-B.6.06	en
heal.publicationDate	2003	en
heal.abstract	Synthetic Aperture Radar (SAR) images are extensively used for the determination of oil spills in the marine environment, as they are not affected by local weather condition and cloudiness. However, radar backscatter values for oil spills are very similar to backscatter values for very calm sea areas, because dampen capillary and short gravity waves is caused by the presence of an oil spill. Most classification algorithms for oil spill detection rely on statistic-based decisions. In previous study features which lead to a successful oil spill detection were defined and evaluated through an hybrid multi-segmentation - expert system procedure. In this paper a classification scheme based on neural networks is presented. In order the neural network to be functional and the classification procedure to be simple, the inputs of the neural network are images. Thus, several images were generated from the original SAR, each one presenting a texture or geometry key-feature. The potential of the Multilayer Perceptron neural network and different training algorithms for oil spill classification were investigated. As appropriate topology for oil spill detection is specified the 4:2:1 and training algorithm the standard backpropagation for a certain number of epochs and afterwards the conjugate gradient or resilient backpropagation. Methods overall accuracy for the examined areas vary from 99.3% to 99.6%. Copyright © 2003 by the International Astronautical Federation. All rights reserved.	en
heal.journalName	54th International Astronautical Congress of the International Astronautical Federation (IAF), the International Academy of Astronautics and the International Institute of Space Law	en
dc.identifier.volume	1	en
dc.identifier.spage	1053	en
dc.identifier.epage	1060	en