Automated oil spill detection using deep learning and SAR satellite data for the northern entrance of the Suez Canal
Abstract Oil spills threaten marine ecosystems, demanding swift detection and response. The northern entrance of the Suez Canal, a critical maritime route, is increasingly at risk of frequent oil spill incidents. This study employs the DeepLabv3 + deep learning model to automatically detect oil spil...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Nature Portfolio
2025-06-01
|
| Series: | Scientific Reports |
| Subjects: | |
| Online Access: | https://doi.org/10.1038/s41598-025-03028-1 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Abstract Oil spills threaten marine ecosystems, demanding swift detection and response. The northern entrance of the Suez Canal, a critical maritime route, is increasingly at risk of frequent oil spill incidents. This study employs the DeepLabv3 + deep learning model to automatically detect oil spills in the study area based on Sentinel-1 Synthetic Aperture Radar imagery provided by the European Space Agency. The model was trained separately on two datasets: the European Maritime Safety Agency CleanSeaNet (EMSA-CSN) dataset, comprising 1100 oil spill incidents, and a localized dataset containing 1500 oil spill incidents that occurred at the Egyptian territorial waters. A comparative analysis between the two models was conducted using 30 oil spill test cases located within the study area. The model trained on Egyptian data outperformed the EMSA-CSN-data- trained model, achieving a loss of 0.0516, an accuracy of 98.14%, a mean Intersection over Union (MIoU) of 0.7872, and a significantly higher ROC area of 0.91, compared to a loss of 0.1152, an accuracy of 96.45%, a MIoU of 0.7161, and a ROC area of 0.76 for the EMSA-CSN model. In addition, the area prediction analysis confirmed the superior performance of the Egyptian-data-trained model, which estimated a total affected area of 421.20 km2, closely aligning with the ground truth of 425.20 km2, whereas the EMSA-CSN-data-trained model underestimated oil spills of around 323.98 km2. These results highlight the benefits of region-specific training in improving segmentation quality and reducing errors. This study emphasizes the potential of AI-driven models for real-time oil spill monitoring, with applications in environmental protection and emergency response. |
|---|---|
| ISSN: | 2045-2322 |