Scholarlink

<p>Coastal water pollution poses significant threats to marine ecosystems, fisheries, and human well-being, particularly in densely populated regions such as the Recife Coastal Zone, Brazil. This study presents a near real-time framework integrating Sentinel-2 satellite data with Convolutional Neural Networks (CNNs). The system enables automated, rapid detection of pollution immediately after satellite data acquisition, bridging the gap between traditional post-processing and timely environmental monitoring. It also strives for the integration of remote sensing data and CNNs for operational monitoring of water quality. The CNN was trained on 700 and validated on 300 samples using optimized hyperparameters (100 epochs; batch size: 32). Model evaluation yielded R² = 0.91, RMSE = 3.8&#xa0;mg/L, and MAE = 2.5&#xa0;mg/L, while classification achieved 92.4% accuracy, 91.6% precision, 93.1% recall, and an F1-score of 92.3%. These metrics highlight the reliability of the CNN in estimating pollutant concentration and spatial distribution in near real time. The findings demonstrate that deep learning–based frameworks can significantly enhance satellite-driven coastal water quality monitoring, supporting proactive management and mitigation strategies for pollution control in vulnerable coastal regions.</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

Deep Learning-Enabled Real-Time Monitoring of Coastal Water Pollution Using Sentinel-2 Remote Sensing Data

  • 期刊论文

  • 发表日期：

    2026年1月12日

查看全文
  • Eatedal Alabdulkreem,
  • Turke Althobaiti,
  • Mohammed Abaker,
  • G. Murugesan,
  • Hanadi Alkhudhayr,
  • S. Srinivasan,
  • V. Sampathkumar,
  • V. Priya

摘要

Coastal water pollution poses significant threats to marine ecosystems, fisheries, and human well-being, particularly in densely populated regions such as the Recife Coastal Zone, Brazil. This study presents a near real-time framework integrating Sentinel-2 satellite data with Convolutional Neural Networks (CNNs). The system enables automated, rapid detection of pollution immediately after satellite data acquisition, bridging the gap between traditional post-processing and timely environmental monitoring. It also strives for the integration of remote sensing data and CNNs for operational monitoring of water quality. The CNN was trained on 700 and validated on 300 samples using optimized hyperparameters (100 epochs; batch size: 32). Model evaluation yielded R² = 0.91, RMSE = 3.8 mg/L, and MAE = 2.5 mg/L, while classification achieved 92.4% accuracy, 91.6% precision, 93.1% recall, and an F1-score of 92.3%. These metrics highlight the reliability of the CNN in estimating pollutant concentration and spatial distribution in near real time. The findings demonstrate that deep learning–based frameworks can significantly enhance satellite-driven coastal water quality monitoring, supporting proactive management and mitigation strategies for pollution control in vulnerable coastal regions.