You must have a javascript-enabled browser and javacript and stylesheets must be enabled to use some of the functions on this site.
 
   workshops     

 

Neural Networks for Oil Spill Detection Using ERS and ENVISAT Imagery

Fabio Del Frate(1) , Emanuele Angiuli(1) , and Luca Salvatori(1)

(1) Tor Vergata University, Via del Politecnico, 1, 00133 Rome, Italy

Abstract

In this paper we present the potentialities of Neural Network algorithms for detection of oil spills. An artificial neural network (NN) may be viewed as a mathematical model composed of many nonlinear computational elements, named neurons, operating in parallel and massively connected by links characterized by different weights. This particular structure makes neural networks very stable and robust when there are sensible input variations. For this study, multilayer perceptrons (MPL) have been considered, which have been found to have a suited topology for classification and inversion problems. The network input is a vector containing the values of a set of features characterizing an oil spill candidate. The output gives the probability for the candidate to be a real oil spill. The classification performance has been evaluated on a data set of ERS-SAR and Envisat-ASAR images containing examples of oil spills and look-alikes. To test the robustness of the algorithm, we have grouped all the collected examples into different subsets. Each subset contained examples characterized by the same SAR instrument (ERS-SAR or Envisat-ASAR) and/or affected by the same level of natural noise. Several nets have been designed using in turn different subsets for the training phase and the performance of each net has been tested on different subsets, either similar or dissimilar to those used in the training. The neural algorithm gives generally satisfactory results, however the performance is clearly affected by the type of examples included in the training set. This effect has been investigated throughout a systematic analysis.

 

Full paper

 

  Higher level                 Last modified: 07.10.03