OIL SPILL DETECTION AND CHARACTERIZATION USING FULLY-POLARIMETRIC X AND C BAND SAR IMAGERY: A NEAR REAL TIME PERSPECTIVE

摘要

We explores the possibilities and advantages of quad polarimetricSAR data for the purpose of oil spill detectionand discrimination of different types of slicks and lookalikes.An array of polarimetric features derived fromthe Pauli and lexicographic basis scattering matrices havebeen proposed. Those sets of features are then used totrain and validate an Artificial Neural Network classifier.On a dataset of near-coincident TerraSAR-X (TS-X) andRADARSAT-2 (RS-2) acquisitions, we perform a featureanalysis in terms of relevance and redundancy for oil slickcharacterization and ranked according to their ability todiscriminate between oil spills and look-alikes. Polarimetricfeatures such as Scattering diversity, Surface scatteringfraction, Entropy and Span proved to be more discriminativethan other polarimetric features.