Sea fog detection from GK2A AMI and GK2B GOCI-II satellite images using swin transformer

Sea fog detection is a critical aspect of meteorological monitoring due to its significant impact on maritime safety and navigation. However, accurately detecting sea fog poses challenges due to its dynamic nature and the limitations of conventional detection methods. Recent advancements in remote sensing technology and deep learning provide an opportunity to overcome these challenges. This study leverages the capabilities of Korea’s geostationary satellites, GK2A AMI and GK2B GOCI-II, and applies a state-of-the-art deep learning model, Swin Transformer-based UPerNet, to develop an efficient sea fog detection system. To achieve this, satellite images from AMI and GOCI-II were collected, preprocessed, and labeled using manual and automated methods. Composite images, generated from selected spectral bands effective for fog detection, served as inputs to the model. The datasets were augmented and standardized to enhance model performance and generalization. The trained model was evaluated using metrics such as overall accuracy (ACC) and critical success index (CSI), achieving 98.8% ACC and 78.76% CSI, respectively, on the test dataset. The results demonstrate the potential of the proposed approach to improve sea fog detection, with applications extending to operational meteorology and maritime safety. Although limitations such as minor distortions in detection accuracy were observed, these can be addressed in future studies by incorporating more advanced models and additional data sources. This research highlights the synergy between geostationary satellite data and deep learning for environmental monitoring and provides a foundation for further advancements in remote sensing applications.

 

This research was supported by a grant (2021-MOIS37-002) of "Intelligent Technology Development Program on Disaster Response and Emergency Management" funded by Ministry of Interior and Safety (MOIS, Korea).