Dealing with missing modalities at test time for land cover mapping: A case study on multi-source optical data

Yawogan Jean Eudes Gbodjo; Dino Ienco

doi:10.1109/MELECON53508.2022.9842865

Dealing with missing modalities at test time for land cover mapping: A case study on multi-source optical data

Yawogan Jean Eudes Gbodjo
, Dino Ienco

Not published at VTT

National Research Institute for Agriculture, Food and Environment (INRAE)

Research output: Chapter in Book/Report/Conference proceeding › Conference article in proceedings › Scientific › peer-review

Abstract

In recent years, multiple sources of remote sensing data have become increasingly available to monitor Earth's surface phenomena. However, unlike High Spatial Resolution (HSR) data, Very High Spatial Resolution (VHSR) satellite images remain difficult to collect over large areas due to acquisition costs and a smaller swath. This often compromises the simultaneous use of both sources of data over same study areas for many applications. In this work, we investigate a land cover mapping setting in which both HSR and VHSR are available at the learning stage of a deep neural network while only the HSR data is available at inference time for model inference. We thus propose simple but effective strategies for enhancing the land cover classification in this scenario of incomplete multi-source remote sensing data when the model is deployed.

Original language	English
Title of host publication	2022 IEEE 21st Mediterranean Electrotechnical Conference (MELECON)
Publisher	IEEE Institute of Electrical and Electronic Engineers
Pages	621-626
Number of pages	6
ISBN (Electronic)	978-1-6654-4280-0
DOIs	https://doi.org/10.1109/MELECON53508.2022.9842865
Publication status	Published - 2022
MoE publication type	A4 Article in a conference publication
Event	21st IEEE Mediterranean Electrotechnical Conference, MELECON 2022 - Palermo, Italy Duration: 14 Jun 2022 → 16 Jun 2022

Conference

Conference	21st IEEE Mediterranean Electrotechnical Conference, MELECON 2022
Country/Territory	Italy
City	Palermo
Period	14/06/22 → 16/06/22

Funding

This work was supported by the Programme National de Télédétection Spatiale under Grant PNTS-2020-13.

Keywords

deep neural networks
high and very high spatial resolution
land use land cover mapping
missing modalities
Multi-source remote sensing

Access to Document

10.1109/MELECON53508.2022.9842865

Cite this

@inproceedings{6c07c6785f9449d3a163af94f751a015,

title = "Dealing with missing modalities at test time for land cover mapping: A case study on multi-source optical data",

abstract = "In recent years, multiple sources of remote sensing data have become increasingly available to monitor Earth's surface phenomena. However, unlike High Spatial Resolution (HSR) data, Very High Spatial Resolution (VHSR) satellite images remain difficult to collect over large areas due to acquisition costs and a smaller swath. This often compromises the simultaneous use of both sources of data over same study areas for many applications. In this work, we investigate a land cover mapping setting in which both HSR and VHSR are available at the learning stage of a deep neural network while only the HSR data is available at inference time for model inference. We thus propose simple but effective strategies for enhancing the land cover classification in this scenario of incomplete multi-source remote sensing data when the model is deployed.",

keywords = "deep neural networks, high and very high spatial resolution, land use land cover mapping, missing modalities, Multi-source remote sensing",

author = "Gbodjo, \{Yawogan Jean Eudes\} and Dino Ienco",

year = "2022",

doi = "10.1109/MELECON53508.2022.9842865",

language = "English",

pages = "621--626",

booktitle = "2022 IEEE 21st Mediterranean Electrotechnical Conference (MELECON)",

publisher = "IEEE Institute of Electrical and Electronic Engineers",

address = "United States",

note = "21st IEEE Mediterranean Electrotechnical Conference, MELECON 2022 ; Conference date: 14-06-2022 Through 16-06-2022",

}

Gbodjo, YJE & Ienco, D 2022, Dealing with missing modalities at test time for land cover mapping: A case study on multi-source optical data. in 2022 IEEE 21st Mediterranean Electrotechnical Conference (MELECON). IEEE Institute of Electrical and Electronic Engineers, pp. 621-626, 21st IEEE Mediterranean Electrotechnical Conference, MELECON 2022, Palermo, Italy, 14/06/22. https://doi.org/10.1109/MELECON53508.2022.9842865

Dealing with missing modalities at test time for land cover mapping: A case study on multi-source optical data. / Gbodjo, Yawogan Jean Eudes; Ienco, Dino.
2022 IEEE 21st Mediterranean Electrotechnical Conference (MELECON). IEEE Institute of Electrical and Electronic Engineers, 2022. p. 621-626.

Research output: Chapter in Book/Report/Conference proceeding › Conference article in proceedings › Scientific › peer-review

TY - GEN

T1 - Dealing with missing modalities at test time for land cover mapping

T2 - 21st IEEE Mediterranean Electrotechnical Conference, MELECON 2022

AU - Gbodjo, Yawogan Jean Eudes

AU - Ienco, Dino

PY - 2022

Y1 - 2022

N2 - In recent years, multiple sources of remote sensing data have become increasingly available to monitor Earth's surface phenomena. However, unlike High Spatial Resolution (HSR) data, Very High Spatial Resolution (VHSR) satellite images remain difficult to collect over large areas due to acquisition costs and a smaller swath. This often compromises the simultaneous use of both sources of data over same study areas for many applications. In this work, we investigate a land cover mapping setting in which both HSR and VHSR are available at the learning stage of a deep neural network while only the HSR data is available at inference time for model inference. We thus propose simple but effective strategies for enhancing the land cover classification in this scenario of incomplete multi-source remote sensing data when the model is deployed.

AB - In recent years, multiple sources of remote sensing data have become increasingly available to monitor Earth's surface phenomena. However, unlike High Spatial Resolution (HSR) data, Very High Spatial Resolution (VHSR) satellite images remain difficult to collect over large areas due to acquisition costs and a smaller swath. This often compromises the simultaneous use of both sources of data over same study areas for many applications. In this work, we investigate a land cover mapping setting in which both HSR and VHSR are available at the learning stage of a deep neural network while only the HSR data is available at inference time for model inference. We thus propose simple but effective strategies for enhancing the land cover classification in this scenario of incomplete multi-source remote sensing data when the model is deployed.

KW - deep neural networks

KW - high and very high spatial resolution

KW - land use land cover mapping

KW - missing modalities

KW - Multi-source remote sensing

UR - https://www.scopus.com/pages/publications/85136439288

U2 - 10.1109/MELECON53508.2022.9842865

DO - 10.1109/MELECON53508.2022.9842865

M3 - Conference article in proceedings

AN - SCOPUS:85136439288

SP - 621

EP - 626

BT - 2022 IEEE 21st Mediterranean Electrotechnical Conference (MELECON)

PB - IEEE Institute of Electrical and Electronic Engineers

Y2 - 14 June 2022 through 16 June 2022

ER -

Dealing with missing modalities at test time for land cover mapping: A case study on multi-source optical data

Abstract

Conference

Funding

Keywords

Access to Document

Other files and links

Fingerprint

Cite this