Scholarlink

<p>Oil pollution poses a significant environmental threat to oceanic and coastal ecosystems, with severe consequences for marine biodiversity, water quality, and local economies. The recent sinking of the MT Terranova tanker in Manila Bay (25 July 2024) released 1.5 million liters of industrial oil into the sea, exacerbating ongoing spill risks. Previous large-scale incidents (e.g., Deepwater Horizon, 2010; Exxon Valdez, 1989) have demonstrated catastrophic ecological and economic impacts, underscoring the need for rapid detection and monitoring. In this work, we propose an end-to-end oil spill monitoring system, "OilSpillNets". Our approach employs a late-fusion (ensemble) strategy that integrates heterogeneous satellite imagery with high-resolution drone data. By capturing large-scale spill patterns via Synthetic Aperture Radar (SAR) and finer details (including oil thickness estimation) from drone imagery, the system provides a comprehensive view of the extent and characteristics of the slick. Lightweight deep learning models are trained on this multimodal dataset and optimised for edge deployment. Evaluation on low-power platforms (Intel Movidius Myriad X and Rockchip RK3588S) demonstrates real-time performance, with the RK3588S achieving &#xa0;76 FPS at a power consumption of &#xa0;3.2 W. These results indicate that our framework can be deployed cost-effectively on satellites, drones, or surface vessels to provide timely early warnings and support cleanup efforts.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Oilspillnets: detecting and estimating oil spills using fusion

  • 期刊论文

  • 发表日期：

    2025年8月18日

查看全文
  • Besma Guesmi,
  • David Moloney

摘要

Oil pollution poses a significant environmental threat to oceanic and coastal ecosystems, with severe consequences for marine biodiversity, water quality, and local economies. The recent sinking of the MT Terranova tanker in Manila Bay (25 July 2024) released 1.5 million liters of industrial oil into the sea, exacerbating ongoing spill risks. Previous large-scale incidents (e.g., Deepwater Horizon, 2010; Exxon Valdez, 1989) have demonstrated catastrophic ecological and economic impacts, underscoring the need for rapid detection and monitoring. In this work, we propose an end-to-end oil spill monitoring system, "OilSpillNets". Our approach employs a late-fusion (ensemble) strategy that integrates heterogeneous satellite imagery with high-resolution drone data. By capturing large-scale spill patterns via Synthetic Aperture Radar (SAR) and finer details (including oil thickness estimation) from drone imagery, the system provides a comprehensive view of the extent and characteristics of the slick. Lightweight deep learning models are trained on this multimodal dataset and optimised for edge deployment. Evaluation on low-power platforms (Intel Movidius Myriad X and Rockchip RK3588S) demonstrates real-time performance, with the RK3588S achieving  76 FPS at a power consumption of  3.2 W. These results indicate that our framework can be deployed cost-effectively on satellites, drones, or surface vessels to provide timely early warnings and support cleanup efforts.