Deep learning-based hyperspectral oil spill detection for marine pollution monitoring in the Gulf of Mexico: A step toward marine pollution monitoring and SDG 14 compliance.

Oil spills pose a significant threat to marine ecosystems, undermining global efforts toward sustainable development goals (SDGs), particularly in preserving marine resources, oceans. This study investigates the application of hyperspectral remote sensing data combined with Deep Learning (DL) techniques to enhance oil spill detection accuracy and efficiency. The idea of this research is to exploit the rich spectral information of hyperspectral imagery together with powerful DL segmentation models to achieve reliable detection of oil-contaminated regions. Using the publicly available Hyperspectral Oil Spill Database (HOSD), we applied Principal Component Analysis (PCA) for dimensionality reduction and implemented standardized patch-based input preparation to address both data scarcity and high dimensionality. Two convolutional neural network architectures, U-Net and DeepLabv3, were employed for semantic segmentation of oil-contaminated regions. Model performance was evaluated using Intersection over Union (IoU), F1 score, precision, and recall. DeepLabv3 achieved higher precision and greater IoU (57.86 %) compared to U-Net, while U-Net demonstrated superior sensitivity. These results highlight the complementary strengths of both models and underscore the potential of combining hyperspectral imaging and DL for effective oil spill detection, contributing to improved environmental monitoring in alignment with the sustainable development goals.