研究目的
Investigating the use of deep learning based architectures for automated eddy detection and classification from Sea Surface Height (SSH) maps.
研究成果
EddyNet successfully applies deep learning techniques to the problem of eddy classification, achieving comparable results with a smaller number of parameters and faster training times. Future work could include the application of EddyNet globally and the addition of other surface information to improve detection.
研究不足
The study is limited to the Southern Atlantic Ocean region and uses a specific resolution of SSH maps (0.25°). The architecture may overfit due to the low number of training samples compared to the capacity of the architecture.
1：Experimental Design and Method Selection:
The study employs a convolutional encoder-decoder architecture inspired by U-Net for pixel-wise classification of oceanic eddies.
2：Sample Selection and Data Sources:
Uses 15 years (1998-2012) of daily detected and classified eddies from the Southern Atlantic Ocean region, provided by the Copernicus Marine Environment Monitoring Service (CMEMS).
3：List of Experimental Equipment and Materials:
Utilizes a Nvidia K80 GPU card for training the neural network.
4：Experimental Procedures and Operational Workflow:
The training dataset is split into training and validation sets, with an early-stopping strategy to prevent overfitting.
5：Data Analysis Methods:
Performance is assessed using the Dice coefficient and global accuracy metrics.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
