Accounting for label errors when training a convolutional neural network to estimate sea ice concentration using operational ice charts
Convolutional neural networks (CNNs) are being increasingly investigated as a means to extract sea ice concentration from SAR in an automated manner. Often this is done using ice charts as training data. However, in these charts an ice concentration label is given to a large region, which may not ha...
Published in: | IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing |
---|---|
Main Authors: | , , |
Format: | Article in Journal/Newspaper |
Language: | English |
Published: |
2022
|
Subjects: | |
Online Access: | https://orbit.dtu.dk/en/publications/34900a9f-d9c2-4cab-aa10-13845b8552c8 https://doi.org/10.1109/JSTARS.2022.3141063 https://backend.orbit.dtu.dk/ws/files/270485826/Accounting_for_Label_Errors_When_Training_a_Convolutional_Neural_Network_to_Estimate_Sea_Ice_Concentration_Using_Operational_Ice_Charts.pdf |
id |
ftdtupubl:oai:pure.atira.dk:publications/34900a9f-d9c2-4cab-aa10-13845b8552c8 |
---|---|
record_format |
openpolar |
spelling |
ftdtupubl:oai:pure.atira.dk:publications/34900a9f-d9c2-4cab-aa10-13845b8552c8 2024-09-15T18:34:15+00:00 Accounting for label errors when training a convolutional neural network to estimate sea ice concentration using operational ice charts Tamber, Manveer Scott, K. Andrea Pedersen, Leif Toudal 2022 application/pdf https://orbit.dtu.dk/en/publications/34900a9f-d9c2-4cab-aa10-13845b8552c8 https://doi.org/10.1109/JSTARS.2022.3141063 https://backend.orbit.dtu.dk/ws/files/270485826/Accounting_for_Label_Errors_When_Training_a_Convolutional_Neural_Network_to_Estimate_Sea_Ice_Concentration_Using_Operational_Ice_Charts.pdf eng eng https://orbit.dtu.dk/en/publications/34900a9f-d9c2-4cab-aa10-13845b8552c8 info:eu-repo/semantics/openAccess Tamber , M , Scott , K A & Pedersen , L T 2022 , ' Accounting for label errors when training a convolutional neural network to estimate sea ice concentration using operational ice charts ' , IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing , vol. 15 , pp. 1502-1513 . https://doi.org/10.1109/JSTARS.2022.3141063 Convolutional neural network (CNN) Ice Concentration Synthetic Aperture Radar (SAR) data article 2022 ftdtupubl https://doi.org/10.1109/JSTARS.2022.3141063 2024-08-13T00:03:06Z Convolutional neural networks (CNNs) are being increasingly investigated as a means to extract sea ice concentration from SAR in an automated manner. Often this is done using ice charts as training data. However, in these charts an ice concentration label is given to a large region, which may not have a spatially uniform sea ice concentration distribution at the prediction scale of the CNN. This leads to representativity errors, which can be more pronounced at intermediate sea ice concentrations. In this study we first investigate ways to perturb the ice chart labels to obtain improved predictions to account for the label uncertainty for intermediate ice concentrations. We then propose a method to augment the ice chart data by rescaling the information in the SAR imagery. The method is found to lead to improved accuracy in comparison to using the ice chart labels alone, with accuracy improving from 0.921 to 0.979. The sea ice concentration maps with the augmented labels also have much finer detail than the other approaches evaluated. These details are visually in agreement with expected sea ice concentration from the SAR data. Article in Journal/Newspaper Sea ice Technical University of Denmark: DTU Orbit IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 15 1502 1513 |
institution |
Open Polar |
collection |
Technical University of Denmark: DTU Orbit |
op_collection_id |
ftdtupubl |
language |
English |
topic |
Convolutional neural network (CNN) Ice Concentration Synthetic Aperture Radar (SAR) data |
spellingShingle |
Convolutional neural network (CNN) Ice Concentration Synthetic Aperture Radar (SAR) data Tamber, Manveer Scott, K. Andrea Pedersen, Leif Toudal Accounting for label errors when training a convolutional neural network to estimate sea ice concentration using operational ice charts |
topic_facet |
Convolutional neural network (CNN) Ice Concentration Synthetic Aperture Radar (SAR) data |
description |
Convolutional neural networks (CNNs) are being increasingly investigated as a means to extract sea ice concentration from SAR in an automated manner. Often this is done using ice charts as training data. However, in these charts an ice concentration label is given to a large region, which may not have a spatially uniform sea ice concentration distribution at the prediction scale of the CNN. This leads to representativity errors, which can be more pronounced at intermediate sea ice concentrations. In this study we first investigate ways to perturb the ice chart labels to obtain improved predictions to account for the label uncertainty for intermediate ice concentrations. We then propose a method to augment the ice chart data by rescaling the information in the SAR imagery. The method is found to lead to improved accuracy in comparison to using the ice chart labels alone, with accuracy improving from 0.921 to 0.979. The sea ice concentration maps with the augmented labels also have much finer detail than the other approaches evaluated. These details are visually in agreement with expected sea ice concentration from the SAR data. |
format |
Article in Journal/Newspaper |
author |
Tamber, Manveer Scott, K. Andrea Pedersen, Leif Toudal |
author_facet |
Tamber, Manveer Scott, K. Andrea Pedersen, Leif Toudal |
author_sort |
Tamber, Manveer |
title |
Accounting for label errors when training a convolutional neural network to estimate sea ice concentration using operational ice charts |
title_short |
Accounting for label errors when training a convolutional neural network to estimate sea ice concentration using operational ice charts |
title_full |
Accounting for label errors when training a convolutional neural network to estimate sea ice concentration using operational ice charts |
title_fullStr |
Accounting for label errors when training a convolutional neural network to estimate sea ice concentration using operational ice charts |
title_full_unstemmed |
Accounting for label errors when training a convolutional neural network to estimate sea ice concentration using operational ice charts |
title_sort |
accounting for label errors when training a convolutional neural network to estimate sea ice concentration using operational ice charts |
publishDate |
2022 |
url |
https://orbit.dtu.dk/en/publications/34900a9f-d9c2-4cab-aa10-13845b8552c8 https://doi.org/10.1109/JSTARS.2022.3141063 https://backend.orbit.dtu.dk/ws/files/270485826/Accounting_for_Label_Errors_When_Training_a_Convolutional_Neural_Network_to_Estimate_Sea_Ice_Concentration_Using_Operational_Ice_Charts.pdf |
genre |
Sea ice |
genre_facet |
Sea ice |
op_source |
Tamber , M , Scott , K A & Pedersen , L T 2022 , ' Accounting for label errors when training a convolutional neural network to estimate sea ice concentration using operational ice charts ' , IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing , vol. 15 , pp. 1502-1513 . https://doi.org/10.1109/JSTARS.2022.3141063 |
op_relation |
https://orbit.dtu.dk/en/publications/34900a9f-d9c2-4cab-aa10-13845b8552c8 |
op_rights |
info:eu-repo/semantics/openAccess |
op_doi |
https://doi.org/10.1109/JSTARS.2022.3141063 |
container_title |
IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing |
container_volume |
15 |
container_start_page |
1502 |
op_container_end_page |
1513 |
_version_ |
1810476057253380096 |