Landsat Ice Classification for the Baltic Sea: Development of a near real time service using neural networks
Artificial neural networks show a promising potential for automatized sea ice classification services. The DLR - Maritime Safety and Security Lab has been developing a deep net in collaboration with the German Ice Service to provide near real time mapping of the stage of development for the Baltic s...
Main Authors: | , |
---|---|
Format: | Conference Object |
Language: | English |
Published: |
2023
|
Subjects: | |
Online Access: | https://elib.dlr.de/199447/ https://elib.dlr.de/199447/1/IICWG_2023_Kowalewski_DLR.pdf https://nsidc.org/noaa/iicwg/iicwg-meetings |
Summary: | Artificial neural networks show a promising potential for automatized sea ice classification services. The DLR - Maritime Safety and Security Lab has been developing a deep net in collaboration with the German Ice Service to provide near real time mapping of the stage of development for the Baltic sea ice. The classification is based on optical imagery from the Landsat mission and first results indicate a promising mapping performance. However, the current version of our deep net does not consider contextual information from sea ice patches, which would further improve its classification performance. We are currently developing a convolutional neural network, which will be capable of taking the spatial context of sea ice type into consideration. We'll address its integration to the processing infrastructure of the German Earth Observation Centre and potential fusion with additional sensors. |
---|