Please use this identifier to cite or link to this item: http://hdl.handle.net/2080/3970
Full metadata record
DC FieldValueLanguage
dc.contributor.authorSaida, Shaik John-
dc.contributor.authorAri, Samit-
dc.date.accessioned2023-03-09T07:18:36Z-
dc.date.available2023-03-09T07:18:36Z-
dc.date.issued2023-02-
dc.identifier.citation3rd IEEE International Conference On Range Technology(ICORT 2023), Chandipur, India, 23 To 25 February 2023en_US
dc.identifier.urihttp://hdl.handle.net/2080/3970-
dc.descriptionCopyright belongs to proceeding publisheren_US
dc.description.abstractAn eddy is a circular movement of ocean water that carries several ocean elements across the ocean. The study of marine biological environments and climate change benefits from the identification of ocean eddies. Eddy detection is one of the most active fields in physical oceanography research. Despite recent advancements, deep convolution neural networks for eddy detection are still in their infancy. The wide range of eddies’ sizes and shapes makes automatic eddy segmentation challenging. Through the creation of a dense forecast, UNet solves this problem. The network architecture, however, is immensely intricate. In this study, a dilated convolution U-Net is proposed for the semantic segmentation of ocean eddies. This method reduces architectural complexity without compromising performance. Further, a new attention approach is proposed to enhance the decoder features before cascading with the encoded feature map. The experimental results demonstrate that the proposed architecture outperforms the state-of-the art techniques for eddy detectionen_US
dc.subjectAutomatic eddy detectionen_US
dc.subjectdeep learningen_US
dc.subjectdilated U-Neten_US
dc.subjectnew attention moduleen_US
dc.titleAttention Aware U-Net Architecture for Semantic Segmentation and Detection of Ocean Eddy’sen_US
dc.typeArticleen_US
Appears in Collections:Conference Papers

Files in This Item:
File Description SizeFormat 
2023_ICORT_SJSaida_Attention.pdf1.55 MBAdobe PDFView/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.