Eddy Detection in the Marginal Ice Zone with Sentinel-1 Data Using YOLOv5

Abstract

The automatic detection and analysis of ocean eddies in the marginal ice zone via remote sensing is a very challenging task but of critical importance for scientific applications and anthropogenic activities. Therefore, as one of the first steps toward the automation of the eddy detection process, we investigated the potential of applying YOLOv5, a deep convolutional neural network architecture, to specifically collected and labeled high-resolution synthetic aperture radar data for a very dynamic area over the Fram Strait. Our approach involved fine-tuning pre-trained YOLOv5 models on a sparse dataset and achieved accurate results with minimal training data. The performances of the models were evaluated using several metrics, and the best model was selected by visual examination. The experimental results obtained from the validation and test datasets consistently demonstrated the robustness and effectiveness of the chosen model to identify submesoscale and mesoscale eddies with different structures. Moreover, our work provides a foundation for automated eddy detection in the marginal ice zone using synthetic aperture radar imagery and contributes to advancing oceanography research.

EddyDet: A Deep Framework for Oceanic Eddy Detection in Synthetic Aperture Radar Images

“…In recent years, an increasing number of studies have been conducted on the detection of eddies using SAR data [17,27,33,[44][45][46]]. Among them, several studies have utilized deep learning approaches to automatically detect eddies in SAR images.…”

“…It is noteworthy that the MFNN model does not provide identification for individual instances of eddies. Xia et al [17] and Khachatrian et al [45] employed YOLO-based networks [48] to detect the bounding boxes of eddies. However, compared to pixel-wise masks that provide detailed information for each object instance, the limited information provided by bounding boxes may not be sufficient for various downstream tasks, such as eddy parameter inversion.…”