Mesoscale Ocean Eddy Detection Using High-Resolution Network

Abstract

Ocean eddies are common phenomena of ocean water movement. They have a significant impact on the physical properties of marine hydrology, marine chemistry and marine biological environment. Current study of ocean eddy detection has already become one of the most active research areas in physical oceanography. Recent trend in eddy detection attempts to employ deep learning methods, but it is still in the early stage. Accordingly, this work takes the advantage of the rapid development of deep learning to improve the current result on ocean eddy detection. We apply the improved and reliable high-resolution representation network to eddy detection and classification from Sea Surface Height (SSH) maps based on semantic segmentation. This high-resolution network can aggregate representations from all the parallel convolutions and repeat the operation of feature fusion. It can therefore maintain and eventually produce high-resolution representations throughout the whole feature extraction process. We then effectively combine the segmentation result with a CascadePSP module and obtain more accurate results than those produced by existing approaches. Our work shows a good performance based on the sea surface height data, which also verifies the application value of deep learning technology in the field of ocean monitoring and data mining.