Creating cloud-free satellite imagery from image time series with deep learning

Publikation: Bidrag til bog/antologi/rapportKonferencebidrag i proceedingsForskningfagfællebedømt

Standard

Creating cloud-free satellite imagery from image time series with deep learning. / Oehmcke, Stefan; Chen, Tzu Hsin Karen; Prishchepov, Alexander V.; Gieseke, Fabian.

Proceedings of the 9th ACM SIGSPATIAL International Workshop on Analytics for Big Geospatial Data, BIGSPATIAL 2020. red. / Varun Chandola; Ranga Raju Vatsavai; Ashwin Shashidharan. Association for Computing Machinery, 2020. 3.

Publikation: Bidrag til bog/antologi/rapportKonferencebidrag i proceedingsForskningfagfællebedømt

Harvard

Oehmcke, S, Chen, THK, Prishchepov, AV & Gieseke, F 2020, Creating cloud-free satellite imagery from image time series with deep learning. i V Chandola, RR Vatsavai & A Shashidharan (red), Proceedings of the 9th ACM SIGSPATIAL International Workshop on Analytics for Big Geospatial Data, BIGSPATIAL 2020., 3, Association for Computing Machinery, 9th ACM SIGSPATIAL International Workshop on Analytics for Big Geospatial Data - BIGSPATIAL 2020, San Diego / Virtuel, USA, 03/11/2020. https://doi.org/10.1145/3423336.3429345

APA

Oehmcke, S., Chen, T. H. K., Prishchepov, A. V., & Gieseke, F. (2020). Creating cloud-free satellite imagery from image time series with deep learning. I V. Chandola, R. R. Vatsavai, & A. Shashidharan (red.), Proceedings of the 9th ACM SIGSPATIAL International Workshop on Analytics for Big Geospatial Data, BIGSPATIAL 2020 [3] Association for Computing Machinery. https://doi.org/10.1145/3423336.3429345

Vancouver

Oehmcke S, Chen THK, Prishchepov AV, Gieseke F. Creating cloud-free satellite imagery from image time series with deep learning. I Chandola V, Vatsavai RR, Shashidharan A, red., Proceedings of the 9th ACM SIGSPATIAL International Workshop on Analytics for Big Geospatial Data, BIGSPATIAL 2020. Association for Computing Machinery. 2020. 3 https://doi.org/10.1145/3423336.3429345

Author

Oehmcke, Stefan ; Chen, Tzu Hsin Karen ; Prishchepov, Alexander V. ; Gieseke, Fabian. / Creating cloud-free satellite imagery from image time series with deep learning. Proceedings of the 9th ACM SIGSPATIAL International Workshop on Analytics for Big Geospatial Data, BIGSPATIAL 2020. red. / Varun Chandola ; Ranga Raju Vatsavai ; Ashwin Shashidharan. Association for Computing Machinery, 2020.

Bibtex

@inproceedings{c62dfd1b82774ff18156f580492456a3,
title = "Creating cloud-free satellite imagery from image time series with deep learning",
abstract = "Optical satellite images are important for environmental monitoring. Unfortunately, such images are often affected by distortions, such as clouds, shadows, or missing data. This work proposes a deep learning approach for cleaning and imputing satellite images, which could serve as a reliable preprocessing step for spatial and spatio-temporal analyzes. More specifically, a coherent and cloud-free image for a specific target date and region is created based on a sequence of images of that region obtained at previous dates. Our model first extracts information from the previous time steps via a special gating function and then resorts to a modified version of the well-known U-Net architecture to obtain the desired output image. The model uses supplementary data, namely the approximate cloud coverage of input images, the temporal distance to the target time, and a missing data mask for each input time step. During the training phase we condition our model with the targets cloud coverage and missing values (disabled in production), which allows us to use data afflicted by distortion during training and thus does not require pre-selection of distortion-free data. Our experimental evaluation, conducted on data of the Landsat missions, shows that our approach outperforms the commonly utilized approach that resorts to taking the median of cloud-free pixels for a given position. This is especially the case when the quality of the data for the considered period is poor (e.g., lack of cloud free-images during the winter/fall periods). Our deep learning approach allows to improve the utility of the entire Landsat archive, the only existing global medium-resolution free-access satellite archive dating back to the 1970s. It therefore holds scientific and societal potential for future analyses conducted on data from this and other satellite imagery repositories.",
keywords = "image reconstruction, remote sensing, satellite imagery",
author = "Stefan Oehmcke and Chen, {Tzu Hsin Karen} and Prishchepov, {Alexander V.} and Fabian Gieseke",
year = "2020",
month = nov,
day = "3",
doi = "10.1145/3423336.3429345",
language = "English",
editor = "Varun Chandola and Vatsavai, {Ranga Raju} and Ashwin Shashidharan",
booktitle = "Proceedings of the 9th ACM SIGSPATIAL International Workshop on Analytics for Big Geospatial Data, BIGSPATIAL 2020",
publisher = "Association for Computing Machinery",
note = "9th ACM SIGSPATIAL International Workshop on Analytics for Big Geospatial Data - BIGSPATIAL 2020 ; Conference date: 03-11-2020 Through 06-11-2020",

}

RIS

TY - GEN

T1 - Creating cloud-free satellite imagery from image time series with deep learning

AU - Oehmcke, Stefan

AU - Chen, Tzu Hsin Karen

AU - Prishchepov, Alexander V.

AU - Gieseke, Fabian

PY - 2020/11/3

Y1 - 2020/11/3

N2 - Optical satellite images are important for environmental monitoring. Unfortunately, such images are often affected by distortions, such as clouds, shadows, or missing data. This work proposes a deep learning approach for cleaning and imputing satellite images, which could serve as a reliable preprocessing step for spatial and spatio-temporal analyzes. More specifically, a coherent and cloud-free image for a specific target date and region is created based on a sequence of images of that region obtained at previous dates. Our model first extracts information from the previous time steps via a special gating function and then resorts to a modified version of the well-known U-Net architecture to obtain the desired output image. The model uses supplementary data, namely the approximate cloud coverage of input images, the temporal distance to the target time, and a missing data mask for each input time step. During the training phase we condition our model with the targets cloud coverage and missing values (disabled in production), which allows us to use data afflicted by distortion during training and thus does not require pre-selection of distortion-free data. Our experimental evaluation, conducted on data of the Landsat missions, shows that our approach outperforms the commonly utilized approach that resorts to taking the median of cloud-free pixels for a given position. This is especially the case when the quality of the data for the considered period is poor (e.g., lack of cloud free-images during the winter/fall periods). Our deep learning approach allows to improve the utility of the entire Landsat archive, the only existing global medium-resolution free-access satellite archive dating back to the 1970s. It therefore holds scientific and societal potential for future analyses conducted on data from this and other satellite imagery repositories.

AB - Optical satellite images are important for environmental monitoring. Unfortunately, such images are often affected by distortions, such as clouds, shadows, or missing data. This work proposes a deep learning approach for cleaning and imputing satellite images, which could serve as a reliable preprocessing step for spatial and spatio-temporal analyzes. More specifically, a coherent and cloud-free image for a specific target date and region is created based on a sequence of images of that region obtained at previous dates. Our model first extracts information from the previous time steps via a special gating function and then resorts to a modified version of the well-known U-Net architecture to obtain the desired output image. The model uses supplementary data, namely the approximate cloud coverage of input images, the temporal distance to the target time, and a missing data mask for each input time step. During the training phase we condition our model with the targets cloud coverage and missing values (disabled in production), which allows us to use data afflicted by distortion during training and thus does not require pre-selection of distortion-free data. Our experimental evaluation, conducted on data of the Landsat missions, shows that our approach outperforms the commonly utilized approach that resorts to taking the median of cloud-free pixels for a given position. This is especially the case when the quality of the data for the considered period is poor (e.g., lack of cloud free-images during the winter/fall periods). Our deep learning approach allows to improve the utility of the entire Landsat archive, the only existing global medium-resolution free-access satellite archive dating back to the 1970s. It therefore holds scientific and societal potential for future analyses conducted on data from this and other satellite imagery repositories.

KW - image reconstruction

KW - remote sensing

KW - satellite imagery

UR - http://www.scopus.com/inward/record.url?scp=85097582315&partnerID=8YFLogxK

U2 - 10.1145/3423336.3429345

DO - 10.1145/3423336.3429345

M3 - Article in proceedings

AN - SCOPUS:85097582315

BT - Proceedings of the 9th ACM SIGSPATIAL International Workshop on Analytics for Big Geospatial Data, BIGSPATIAL 2020

A2 - Chandola, Varun

A2 - Vatsavai, Ranga Raju

A2 - Shashidharan, Ashwin

PB - Association for Computing Machinery

T2 - 9th ACM SIGSPATIAL International Workshop on Analytics for Big Geospatial Data - BIGSPATIAL 2020

Y2 - 3 November 2020 through 6 November 2020

ER -

ID: 253451768