2:15 PM - 2:30 PM
[ATT35-03] Ocean Internal Wave Monitoring System Based on Visible Light Satellite Imagery and Deep Learning Networks
Keywords:Internal Solitary Waves, Segmentation, Image processing, Deep Learning
In the field of oceanography, Internal Solitary Waves (ISWs) are a magnificent natural phenomenon that lurk beneath the ocean surface. They are ubiquitous in stably stratified ocean environments, with amplitudes reaching up to 240 meters. These ISWs, carrying a large amount of energy, not only significantly affect the underwater navigation of naval submarines and the operation of sonar and weapon systems but also play a crucial role in the complex energy interactions of marine ecosystems. Therefore, accurate localization and tracking of oceanic ISWs have become an important topic in current ocean research.
Previous researchers have primarily used Synthetic Aperture Radar (SAR) to monitor internal waves. However, as an active satellite system, SAR requires the onboard radar system to actively emit pulse signals to the Earth's surface and receive the reflected signal intensity for imaging, consuming a large amount of resources for signal transmission and reception. This poses significant challenges in real-time data acquisition. In contrast, this study utilizes the visible light band of the Japanese meteorological satellite Himawari 8, which has the advantage of obtaining a high-precision image every ten minutes, making the dynamic tracking of internal waves real-time and more cost-effective.
This research proposes a deep learning-based semantic segmentation method that learns spatial and temporal features through a dual-branch architecture to accurately monitor oceanic internal waves. By combining satellite cloud images with precise quartz crystal sensors installed at the seafloor, we have sufficient information to record the passage times of real ISWs and use them as a basis for identifying ISWs in satellite cloud images.
The method proposed in this research enables more accurate and efficient monitoring of oceanic ISWs, providing important data for marine scientific research and engineering. Moreover, this innovative monitoring approach has the potential to be extended to the study of other oceanic phenomena, contributing to the development of marine science.
Previous researchers have primarily used Synthetic Aperture Radar (SAR) to monitor internal waves. However, as an active satellite system, SAR requires the onboard radar system to actively emit pulse signals to the Earth's surface and receive the reflected signal intensity for imaging, consuming a large amount of resources for signal transmission and reception. This poses significant challenges in real-time data acquisition. In contrast, this study utilizes the visible light band of the Japanese meteorological satellite Himawari 8, which has the advantage of obtaining a high-precision image every ten minutes, making the dynamic tracking of internal waves real-time and more cost-effective.
This research proposes a deep learning-based semantic segmentation method that learns spatial and temporal features through a dual-branch architecture to accurately monitor oceanic internal waves. By combining satellite cloud images with precise quartz crystal sensors installed at the seafloor, we have sufficient information to record the passage times of real ISWs and use them as a basis for identifying ISWs in satellite cloud images.
The method proposed in this research enables more accurate and efficient monitoring of oceanic ISWs, providing important data for marine scientific research and engineering. Moreover, this innovative monitoring approach has the potential to be extended to the study of other oceanic phenomena, contributing to the development of marine science.