.png){kind=icon}
.png){kind=icon}
Return
Deep convolution neural network based semantic segmentation for ocean eddy detection
.png){kind=icon}
PRE
AI
DOI：10.1016/j.eswa.2023.119646.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
A circular movement of ocean water known as eddy is crucial for moving various ocean elements across the ocean. They are essential to the biota and circulation of the ocean. The detection of ocean eddies has significant advantages for the research of marine biological habitats and climate change. The evolution of oceanic remote sensing technologies enables their identification in sea surface height images. Deep learning techniques employed for eddy detection are still in their infancy. This paper proposes a deep convolutional neural network model for automated semantic segmentation and identification of ocean eddies at the pixel level. Semantic segmentation requires an understanding of context efficiently for pixel-level recognition. The attention mechanism is proposed to capture contextual information to tackle this challenge. The self-attention process is used in the suggested attention mechanism to define the semantic relationship between any two pixels. A novel module termed the series atrous spatial pyramid module is proposed as an alternative to the multi-scale fusion model currently in use to capture the multi-scale context of feature maps. Further, a new feature enhancing block that cascades encoder outputs with the decoder is also proposed. The experimental findings demonstrate that the proposed architecture has obtained mean pixel accuracy, F-beta score, mean intersection of union score of 94.52%, 94.45%, 88.64% on the Southern Atlantic Ocean dataset and 94.97%, 94.90%, 88.67% on the South China Sea dataset respectively which is better when compared to existing state-of-the-art techniques.
Keywords:
Anti-cyclonic eddy
Cyclonic eddy
Deep convolutional neural network
Semantic segmentation
Ocean eddy detection
Journal
IF:
7.5
Papers:
2.9W
Citations:
10.2W