Improving Oil Spill Detection in Marine Environments Through Deep Learning Approaches

Oil spills represent a major environmental threat, particularly in marine ecosystems. Timely detection and segmentation of oil spills from satellite imagery is crucial for directing rapid response and remediation efforts.This research concludes with an insights by developing a deep learning approach for automated detection and segmentation of oil spills from synthetic aperture radar (SAR) imagery.This research implements and compares three algorithms PSPNET,DeepLapV3 and fully convolutional networks (FCN) with U-Net for semantic segmentation of SAR images.All the networks are trained on a dataset of SAR images with pixel-wise labels indicating oil spill regions. This research also evaluates the segmentation performance using common metrics like intersection-over-union (IoU) and pixel-wise accuracy. Our results indicate that both FCN and U-Net can effectively detect and delineate oil spill regions, with U-Net demonstrating slightly higher IoU scores and with and accuracy of 95%. The deep learning models generalize well to unseen SAR images. Overall, this work demonstrates the feasibility of deep convolutional neural networks for automated analysis of SAR imagery to enable rapid oil spill response at marine environment.