کنفرانس بین المللی فناوری اطلاعات و دانش

صفحه اصلی / شانزدهمین کنفرانس بین المللی فناوری اطلاعات و دانش

Application of Artificial Intelligence and Remote Sensing for Oil Spill Detection

نویسندگان :

َAmir Reza Ziaee¹ Masomeh Azimzadeh² Parvin Ahmadi³

1- Amirkabir University of Technology 2- ICT Research Institute 3- ICT Research Institute

کلمات کلیدی :

Deep learning،Segmentation،Oil spill detection

چکیده :

Recent advancements in deep learning have improved the automation of oil spill detection; however, most previous studies suffer from limited generalization, low recall on fragmented or thin slicks, and vulnerability to speckle noise in SAR imagery. Many of these works rely solely on encoder-based classification or decoder-based segmentation models without exploring synergistic integration. In this work, we develop a hybrid architecture that combines a modified ResNet-101 encoder with an enhanced U-Net decoder to improve boundary localization and spatial feature retention. Our approach integrates a tailored augmentation pipeline and multi-scale skip refinements specifically adapted to speckle noise and low-contrast marine backgrounds. The model was evaluated on a curated Sentinel-1 SAR dataset with pixel-level annotations. Compared to commonly used architectures such as baseline U-Net, DeepLab variants, and ResNet-U-Net combinations, the proposed method achieved comparatively stronger performance in key metrics such as recall and IoU, while maintaining competitive precision. The results indicate that the proposed method, by balancing spatial detail preservation and accurate boundary detection, achieved higher accuracy in identifying oil spills within complex marine backgrounds