Dilated Convolution based U-Net Architecture for Ocean Eddy Detection

Abstract

Ocean eddies have a significant effect on the maritime environment. They are necessary for carrying a variety of ocean traces across the ocean. Eddy detection is one of the most active fields of physical oceanographic research. Although it is a new trend, using deep learning algorithms to find eddies is still in its early stages. The different sizes and shapes of eddies make automatic eddy segmentation challenging. U-Net makes a dense prediction to solve this problem. However, the network architecture is very intricate. In this paper, a dilated convolution U-Net is developed for the semantic segmentation of ocean eddies using sea surface height data. This technique decreases architectural complexity without sacrificing performance. Further, a new residual path is proposed to cascade encoder outputs with the decoder. The experimental results demonstrate that the proposed architecture outperforms the existing deep learning techniques for eddy detection.