Publication

Abstract : Oil spills over ocean surfaces are mainly caused by release or leakage from the carriers while transportation through water, spill of any fuel or oil refuses used in ships, or even dumping oil remains from the land. Such left-overs on the sea surface may have disastrous consequences, especially because of the fact that they take more time to be accessed and cleaned up. The tides make it easier for the oil to break and spread over long distances. The major consequence is that, oil spills left on the ocean surface affect the marine wildlife and may have adverse effects related to climate and weather conditions. One of the most effective ways to deal with this issue is using remote detection and monitoring of the oil spills. The Synthetic Aperture Radar (SAR) is an instrument that can be used to detect oil spills on sea surface. SAR images show oil spills as dark patches which enables their detection. SAR can operate unaffected by weather conditions and even during night time. SAR observations are taken as input to detect oil spills with deep learning model. The segmented images from the model are trained and tested with different learning rates and varied input feed. The total number of epoches and the number of steps per epoch are changed and the resulting segmented images are studied with respect to loss and accuracy. This work is aimed to contribute to future work on oil spill detection based on effective segmentation.