Scholarlink

<p>The expansion of the raft and aquaculture has substantially impacted worldwide seafood production in the 21st century. Aquaculture monitoring serves to identify both the progress of aquaculture development and water quality status while preventing environmental issues related to pollution or habitat destruction. Nowadays, deep learning models using optical and synthetic-aperture-radar remote sensing images makes monitoring and identifying raft aquaculture possible. This study aims to test DL models based on U-shaped and DeepLab architectures for detecting different types of raft aquaculture using optical and SAR data obtained from Sentinel sensors. Various options to optimize both models related to input fusion data, sub-input sizes, and model structure were applied. As a result, the U- shaped models using the Sentinel-2 images have higher performance in detecting rafts and floating rafts than the models using DeepLab and Sentinel-1 data. Optimal models can detect rafts with accuracy higher than 94% and F1 higher than 80%. The models using Sentinel-1 can only recognize integrated rafts, whereas Sentinel-2 models can distinguish both integrated and bamboo ones. The Sentinel-2 has a high level of detail in spatial resolution, allowing for highly accurate identification. On the other hand, Sentinel-1’s SAR technology is very effective in any weather conditions, making it highly important for uninterrupted surveillance and suitable for real-time monitoring. Integrating these data types requires careful consideration of operational and environmental factors to identify drifting fish rafts. The best model was used to track Vietnamese aquacultural zones throughout seasons and years and may be valuable for coastal managers.</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-based detection of raft aquaculture using Sentinel-1 and Sentinel-2 data

  • 期刊论文

  • 发表日期：

    2025年12月4日

查看全文
  • Van Truong Tran,
  • Kinh Bac Dang,
  • Tuan Linh Giang,
  • Thi Ngan Do,
  • Thi Ngoc Dang,
  • Viet Thanh Pham,
  • Vu Viet Quan Du,
  • Cao Huan Nguyen

摘要

The expansion of the raft and aquaculture has substantially impacted worldwide seafood production in the 21st century. Aquaculture monitoring serves to identify both the progress of aquaculture development and water quality status while preventing environmental issues related to pollution or habitat destruction. Nowadays, deep learning models using optical and synthetic-aperture-radar remote sensing images makes monitoring and identifying raft aquaculture possible. This study aims to test DL models based on U-shaped and DeepLab architectures for detecting different types of raft aquaculture using optical and SAR data obtained from Sentinel sensors. Various options to optimize both models related to input fusion data, sub-input sizes, and model structure were applied. As a result, the U- shaped models using the Sentinel-2 images have higher performance in detecting rafts and floating rafts than the models using DeepLab and Sentinel-1 data. Optimal models can detect rafts with accuracy higher than 94% and F1 higher than 80%. The models using Sentinel-1 can only recognize integrated rafts, whereas Sentinel-2 models can distinguish both integrated and bamboo ones. The Sentinel-2 has a high level of detail in spatial resolution, allowing for highly accurate identification. On the other hand, Sentinel-1’s SAR technology is very effective in any weather conditions, making it highly important for uninterrupted surveillance and suitable for real-time monitoring. Integrating these data types requires careful consideration of operational and environmental factors to identify drifting fish rafts. The best model was used to track Vietnamese aquacultural zones throughout seasons and years and may be valuable for coastal managers.