Deep Red-Tide Learning from GOCI Images

Title
Deep Red-Tide Learning from GOCI Images
Author(s)
김수미; 신지선; 백승재; 유주형
KIOST Author(s)
Kim, Soo Mee(김수미)Baek, Seung Jae(백승재)Ryu, Joo Hyung(유주형)
Alternative Author(s)
김수미; 신지선; 백승재; 유주형
Publication Year
2018-11-04
Abstract
GOCI is a geostationary ocean color satellite launched in 2011 and it provides hourly 8-bands spectral images of Northeast Asian region at 8 times a day with spatial resolution of 500 m. From GOCI images, it is possible to observe a broad ocean area for sensing red-tide occurrence, tidal movement changes and ocean disasters. In this study, we proposed a deep neural network model for automatic detection of red-tide occurrence from a GOCI image. We construct a big training dataset with GOCI nLw (normalized water leaving radiance) images and the corresponding red-tide index maps over 7 years (2011 – 2017). The red-tide index map, which indicates where the red-tide was occurred, was obtained by a decision tree based on the different spectral properties according to the chlorophyll content of red-tide species and the turbidity of the water. The decision tree classifies three red-tide types: i) having low chlorophyll, ii) high chlorophyll in Case-1 water, and iii) high chlorophyll in CASE-2 water of Korean sea area. The datasets are trained to extract some specific spectral features related to the red-tide by a deep neutral network model.cean area for sensing red-tide occurrence, tidal movement changes and ocean disasters. In this study, we proposed a deep neural network model for automatic detection of red-tide occurrence from a GOCI image. We construct a big training dataset with GOCI nLw (normalized water leaving radiance) images and the corresponding red-tide index maps over 7 years (2011 – 2017). The red-tide index map, which indicates where the red-tide was occurred, was obtained by a decision tree based on the different spectral properties according to the chlorophyll content of red-tide species and the turbidity of the water. The decision tree classifies three red-tide types: i) having low chlorophyll, ii) high chlorophyll in Case-1 water, and iii) high chlorophyll in CASE-2 water of Korean sea area. The datasets are trained to extract some specific spectral features related to the red-tide by a deep neutral network model.
URI
https://sciwatch.kiost.ac.kr/handle/2020.kiost/22905
Bibliographic Citation
Pan Ocean Remote Sensing Conference, 2018
Publisher
PORSEC
Type
Conference
Language
English
Publisher
PORSEC
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