DEDNet: Offshore Eddy Detection and Location with HF Radar by Deep Learning

Area of Science:

• Oceanography
• Remote Sensing
• Artificial Intelligence

Background:

• Oceanic eddies significantly impact human activities and oceanographic research.
• Traditional eddy detection methods (Okubo-Weiss, vector geometry, winding angles) require expert experience and extensive computation.
• Automated and accurate detection of offshore eddies is crucial for oceanographic studies.

Purpose of the Study:

• To propose a novel deep eddy detection neural network (DEDNet) for automated offshore eddy detection.
• To compare the performance of DEDNet with existing classical methods and a fully convolutional network (FCN) based approach.
• To develop a more efficient and accurate method for oceanic eddy detection using high-frequency radar data.

Main Methods:

• A deep eddy detection neural network (DEDNet) utilizing pixel segmentation on high-frequency radar (HFR) data was developed.
• An offshore eddy detection dataset was constructed using sea surface current data from HFR systems in the South China Sea.
• A spatial globally optimum and detail-distinguishing pixel segmentation network was implemented for automated eddy localization.
• A fully convolutional network (FCN) based eddy detection network was used for comparative analysis.

Main Results:

• DEDNet demonstrated superior performance compared to the FCN-based eddy detection network.
• The proposed DEDNet achieved competitive results when compared to classical statistical methods.
• The study successfully constructed a dataset for offshore eddy detection using HFR data.

Conclusions:

• DEDNet provides an effective and automated solution for offshore eddy detection.
• The deep learning approach offers advantages over traditional methods in terms of accuracy and computational efficiency.
• This research contributes to advancing oceanographic research through improved eddy detection techniques.