Oil Spill Detection using Convolutional Neural Networks and Sentinel-1 SAR Imagery

Abstract

Oil spills impose significant environmental challenges, leading to critical consequences for marine ecosystems and sea habitant’s health. Early delineatin and efficient surveillance are absolutely important to prevent more contamination and support quick hazards reduction. This study focuses on detecting oil spills using satellite imagery and deep learning models, specifically Convolutional Neural Networks (CNN). The dataset used to train the CNN comprised 695 images extracted from Sentinel-1 Synthetic Aperture Radar (SAR) data over the Mediterranean Sea. In particular, 486 images (70%) were allocated for training, 139 images (20%) for validation, and 70 images (10%) for testing. Preprocessing involved a thresholding technique to enhance feature extraction and improve classification precision. The CNN model achieved a high test accuracy of 98.57%, with perfect precision (1.0000), recall of 96.43%, and F1 score of 0.9818, demonstrating strong performance and reliability. These high accuracy levels underscore the model’s efficiency in identifying oil spills and its soundness in handling unseen data. The significance of this work is in using satellite-based deep learning models for scalable and automated oil spill detection, therefore providing a reliable and effective substitute for more traditional monitoring systems. The model may be applied over large oceanic areas by using satellite images, thereby supporting marine ecosystem preservation and enhancing environmental risk management connected with oil pollution.