Kang et al. (2025) Multi-Satellite Image Matching and Deep Learning Segmentation for Detection of Daytime Sea Fog Using GK2A AMI and GK2B GOCI-II

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Remote Sensing
• Year: 2025
• Date: 2025-12-23
• Authors: JongGu Kang, Hiroyuki Miyazaki, Seung Hee Kim, M. Kafatos, Daesun Kim, Jinsoo Kim, Yangwon Lee
• DOI: 10.3390/rs18010034

Research Groups

Not explicitly mentioned in the provided text.

Short Summary

This study aimed to enhance sea fog detection accuracy and reliability by integrating multi-satellite imagery using a deep learning-based co-registration technique and autotuning state-of-the-art semantic segmentation models. The approach, particularly with multi-satellite fusion, significantly improved detection performance, outperforming existing operational products and reducing the omission of disaster-critical information.

Objective

• To achieve higher accuracy and reliability in sea fog detection by employing a deep learning-based advanced co-registration technique for multi-satellite image fusion and autotuning-based optimization of State-of-the-Art (SOTA) semantic segmentation models.

Study Configuration

• Spatial Scale: Regional scale, covering vast marine areas, specifically utilizing geostationary satellites over the Korean peninsula and surrounding waters.
• Temporal Scale: Continuous monitoring, characteristic of geostationary satellite operations.

Methodology and Data

• Models used: Deep learning-based advanced co-registration technique for multi-satellite image fusion, autotuning-based optimization, Swin Transformer, Mask2Former, and SegNeXt (semantic segmentation models).
• Data sources: Advanced Meteorological Imager (AMI) sensor on the Geostationary Korea Multi-Purpose Satellite 2A (GK2A) and GOCI-II sensor on the Geostationary Korea Multi-Purpose Satellite 2B (GK2B).

Main Results

• Swin Transformer achieved an Intersection over Union (IoU) of 77.24% and an F1-score of 87.16%.
• Multi-satellite fusion significantly improved the Recall score from 88.78% (single AMI product) to 92.01%, effectively mitigating the omission of disaster information.
• Swin Transformer, Mask2Former, and SegNeXt demonstrated balanced and excellent performance across overall metrics.
• The developed deep learning approach was superior to both the officially operational GK2A AMI Fog and GK2B GOCI-II Marine Fog (MF) products.

Contributions

• Introduction of a deep learning-based advanced co-registration technique for multi-satellite image fusion to improve sea fog detection.
• Application and autotuning-based optimization of State-of-the-Art (SOTA) semantic segmentation models (Swin Transformer, Mask2Former, SegNeXt) for sea fog detection.
• Demonstrated significant improvement in sea fog detection accuracy and reliability through multi-satellite data fusion, particularly in enhancing Recall.
• Established a superior sea fog detection method compared to existing officially operational satellite products.

Funding

Not explicitly mentioned in the provided text.

Citation

@article{Kang2025MultiSatellite,
  author = {Kang, JongGu and Miyazaki, Hiroyuki and Kim, Seung Hee and Kafatos, M. and Kim, Daesun and Kim, Jinsoo and Lee, Yangwon},
  title = {Multi-Satellite Image Matching and Deep Learning Segmentation for Detection of Daytime Sea Fog Using GK2A AMI and GK2B GOCI-II},
  journal = {Remote Sensing},
  year = {2025},
  doi = {10.3390/rs18010034},
  url = {https://doi.org/10.3390/rs18010034}
}