The scope of the dissertation is to investigate the capabilities of remote sensing techniques in recognition and detection of oil spills on the sea surface using SAR images. Two new methodologies were developed. The first one is based on object-oriented approach and on fuzzy logic. It is fully automated and is differentiated from the existing systems. The second one is based on the use of neural networks. Two neural networks are required to detect and classify oil slicks; one to detect the black formations on the SAR images and one to classify them as oil spills or look-alikes. Furthermore, the statistical characteristics which should be used to distinguish dark formations to oil spills and natural phenomena were investigated. Combinations of 25 characteristics were examined using neural networks and genetic algorithms. The result was the selection of a specific combination of ten characteristics to classify oil spills. This combination was observed in the majority of the examined solutions and it was proved that it can distinguish the oil slicks from the look-alikes in most of the examined cases.

TOPOUZELIS Konstantinos; 

2008-01-10

JRC40315