Sea Fog Classification from GOCI Images using CNN Transfer Learning Models

DC Field Value Language
dc.contributor.author 전호군 -
dc.contributor.author Jonathan Edwin -
dc.contributor.author 김승룡 -
dc.contributor.author 양찬수 -
dc.date.accessioned 2020-07-15T06:34:47Z -
dc.date.available 2020-07-15T06:34:47Z -
dc.date.created 2020-02-11 -
dc.date.issued 2019-10-31 -
dc.identifier.uri https://sciwatch.kiost.ac.kr/handle/2020.kiost/22075 -
dc.description.abstract This study provides an approaching method of classifying sea fog from Geostationary Ocean Color Image, an optical satellite of South Korea. Convolution Neural Network Transfer Learning (CNN-TL) model is used because of a higher classification ability than a single CNN. The CNN-TL model is combined with dataset VGG19 and ResNet50 which have high performance but less layer than other datasets. In classification with 3-bands training images, the CNN-TL shows 96.7% and 93.0% in VGG19 and ResNet50, respectively. On the other hand, only CNN with identical training images shows the accuracy of 85.3% in VGG19 and 52% in VGG19 and ResNet 50. The result can be used to automate local sea fog detection and prediction. -
dc.description.uri 1 -
dc.language English -
dc.publisher IEICE -
dc.relation.isPartOf ICSANE 2019 -
dc.title Sea Fog Classification from GOCI Images using CNN Transfer Learning Models -
dc.type Conference -
dc.citation.conferencePlace JA -
dc.citation.endPage 90 -
dc.citation.startPage 87 -
dc.citation.title ICSANE 2019 -
dc.contributor.alternativeName 전호군 -
dc.contributor.alternativeName Jonathan Edwin -
dc.contributor.alternativeName 김승룡 -
dc.contributor.alternativeName 양찬수 -
dc.identifier.bibliographicCitation ICSANE 2019, pp.87 - 90 -
dc.description.journalClass 1 -
Appears in Collections:
Sea Power Enhancement Research Division > Marine Domain & Security Research Department > 2. Conference Papers
Marine Digital Resources Department > Marine Bigdata & A.I. Center > 2. Conference Papers
Files in This Item:
There are no files associated with this item.

qrcode

Items in ScienceWatch@KIOST are protected by copyright, with all rights reserved, unless otherwise indicated.

Browse