Enhancing semantic segmentation of Ecuadorian shrimp ponds through fine-tuned vision transformers and U-Net architectures utilizing open-source remote sensing data
Aquaculture has emerged as an important pillar of global food production, and shrimp farming plays a critical role in fulfilling the growing demand for seafood. This is especially true in Ecuador, which is recognized as one of the world's largest exporters and producers of shrimp. However, conv...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Taylor & Francis Group
2025-12-01
|
| Series: | International Journal of Digital Earth |
| Subjects: | |
| Online Access: | https://www.tandfonline.com/doi/10.1080/17538947.2025.2538214 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1839612834112077824 |
|---|---|
| author | Daniel Jacome Jianghao Wang Yong Ge |
| author_facet | Daniel Jacome Jianghao Wang Yong Ge |
| author_sort | Daniel Jacome |
| collection | DOAJ |
| description | Aquaculture has emerged as an important pillar of global food production, and shrimp farming plays a critical role in fulfilling the growing demand for seafood. This is especially true in Ecuador, which is recognized as one of the world's largest exporters and producers of shrimp. However, conventional shrimp pond monitoring has limitations owing to the extensive scale and operational complexity. Traditional methods using low-resolution imagery and ground surveys are hampered by cloud cover, outdated maps, and insufficient temporal resolution, leading to inaccurate pond area estimations and hindering timely management. Our framework accurately segmented shrimp ponds from high-resolution satellite images. Using a fine-tuned Prithvi 100M model, we achieved a state-of-the-art mIoU of 0.970 and 0.993 accuracy, respectively. This significantly surpasses other models, such as ViT-base (mIoU = 0.878) and U-Net variants (mIoU = 0.949). The pre-training of the Prithvi 100M model allowed it to effectively capture intricate pond boundaries and subtle internal structures, resulting in highly accurate and detailed segmentation masks. Fine-tuning the encoder proved to be the most effective (mIoU = 0.991), whereas standard data augmentation negatively impacted the performance. This methodology offers a valuable tool for enhancing water resource management and promoting sustainable aquaculture practices in Ecuadorian coastal regions. |
| format | Article |
| id | doaj-art-ee21d6b9c7544d13b3500d942cf35d6a |
| institution | Matheson Library |
| issn | 1753-8947 1753-8955 |
| language | English |
| publishDate | 2025-12-01 |
| publisher | Taylor & Francis Group |
| record_format | Article |
| series | International Journal of Digital Earth |
| spelling | doaj-art-ee21d6b9c7544d13b3500d942cf35d6a2025-07-28T00:00:49ZengTaylor & Francis GroupInternational Journal of Digital Earth1753-89471753-89552025-12-0118110.1080/17538947.2025.2538214Enhancing semantic segmentation of Ecuadorian shrimp ponds through fine-tuned vision transformers and U-Net architectures utilizing open-source remote sensing dataDaniel Jacome0Jianghao Wang1Yong Ge2State Key Laboratory of Resources and Environmental Information System, Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing, People’s Republic of ChinaState Key Laboratory of Resources and Environmental Information System, Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing, People’s Republic of ChinaState Key Laboratory of Resources and Environmental Information System, Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing, People’s Republic of ChinaAquaculture has emerged as an important pillar of global food production, and shrimp farming plays a critical role in fulfilling the growing demand for seafood. This is especially true in Ecuador, which is recognized as one of the world's largest exporters and producers of shrimp. However, conventional shrimp pond monitoring has limitations owing to the extensive scale and operational complexity. Traditional methods using low-resolution imagery and ground surveys are hampered by cloud cover, outdated maps, and insufficient temporal resolution, leading to inaccurate pond area estimations and hindering timely management. Our framework accurately segmented shrimp ponds from high-resolution satellite images. Using a fine-tuned Prithvi 100M model, we achieved a state-of-the-art mIoU of 0.970 and 0.993 accuracy, respectively. This significantly surpasses other models, such as ViT-base (mIoU = 0.878) and U-Net variants (mIoU = 0.949). The pre-training of the Prithvi 100M model allowed it to effectively capture intricate pond boundaries and subtle internal structures, resulting in highly accurate and detailed segmentation masks. Fine-tuning the encoder proved to be the most effective (mIoU = 0.991), whereas standard data augmentation negatively impacted the performance. This methodology offers a valuable tool for enhancing water resource management and promoting sustainable aquaculture practices in Ecuadorian coastal regions.https://www.tandfonline.com/doi/10.1080/17538947.2025.2538214Pond segmentationremote sensingViTU-Netfine-tuning |
| spellingShingle | Daniel Jacome Jianghao Wang Yong Ge Enhancing semantic segmentation of Ecuadorian shrimp ponds through fine-tuned vision transformers and U-Net architectures utilizing open-source remote sensing data International Journal of Digital Earth Pond segmentation remote sensing ViT U-Net fine-tuning |
| title | Enhancing semantic segmentation of Ecuadorian shrimp ponds through fine-tuned vision transformers and U-Net architectures utilizing open-source remote sensing data |
| title_full | Enhancing semantic segmentation of Ecuadorian shrimp ponds through fine-tuned vision transformers and U-Net architectures utilizing open-source remote sensing data |
| title_fullStr | Enhancing semantic segmentation of Ecuadorian shrimp ponds through fine-tuned vision transformers and U-Net architectures utilizing open-source remote sensing data |
| title_full_unstemmed | Enhancing semantic segmentation of Ecuadorian shrimp ponds through fine-tuned vision transformers and U-Net architectures utilizing open-source remote sensing data |
| title_short | Enhancing semantic segmentation of Ecuadorian shrimp ponds through fine-tuned vision transformers and U-Net architectures utilizing open-source remote sensing data |
| title_sort | enhancing semantic segmentation of ecuadorian shrimp ponds through fine tuned vision transformers and u net architectures utilizing open source remote sensing data |
| topic | Pond segmentation remote sensing ViT U-Net fine-tuning |
| url | https://www.tandfonline.com/doi/10.1080/17538947.2025.2538214 |
| work_keys_str_mv | AT danieljacome enhancingsemanticsegmentationofecuadorianshrimppondsthroughfinetunedvisiontransformersandunetarchitecturesutilizingopensourceremotesensingdata AT jianghaowang enhancingsemanticsegmentationofecuadorianshrimppondsthroughfinetunedvisiontransformersandunetarchitecturesutilizingopensourceremotesensingdata AT yongge enhancingsemanticsegmentationofecuadorianshrimppondsthroughfinetunedvisiontransformersandunetarchitecturesutilizingopensourceremotesensingdata |