Investigating 2d and 3d convolutions for multitemporal land cover classification using remote sensing images

Research output: Contribution to journalConference articleResearchpeer review

Authors

  • M. Voelsen
  • M. Teimouri
  • F. Rottensteiner
  • C. Heipke

External Research Organisations

  • K.N. Toosi University of Technology
View graph of relations

Details

Original languageEnglish
Pages (from-to)271-279
Number of pages9
JournalISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Volume5
Issue number3
Publication statusPublished - 17 May 2022
Event2022 24th ISPRS Congress on Imaging Today, Foreseeing Tomorrow, Commission III - Nice, France
Duration: 6 Jun 202211 Jun 2022

Abstract

With the availability of large amounts of satellite image time series (SITS), the identification of different materials of the Earth's surface is possible with a high temporal resolution. One of the basic tasks is the pixel-wise classification of land cover, i.e.The task of identifying the physical material of the Earth's surface in an image. Fully convolutional neural networks (FCN) are successfully used for this task. In this paper, we investigate different FCN variants, using different methods for the computation of spatial, spectral, and temporal features. We investigate the impact of 3D convolutions in the spatial-Temporal as well as in the spatial-spectral dimensions in comparison to 2D convolutions in the spatial dimensions only. Additionally, we introduce a new method to generate multitemporal input patches by using time intervals instead of fixed acquisition dates. We then choose the image that is closest in time to the middle of the corresponding time interval, which makes our approach more flexible with respect to the requirements for the acquisition of new data. Using these multi-Temporal input patches, generated from Sentinel-2 images, we improve the classification of land cover by 4% in the mean F1-score and 1.3% in the overall accuracy compared to a classification using mono-Temporal input patches. Furthermore, the usage of 3D convolutions instead of 2D convolutions improves the classification performance by a small amount of 0.4% in the mean F1-score and 1.2% in the overall accuracy.

Keywords

    3D-CNN, FCN, land cover classification, Multi-Temporal images, Remote sensing

ASJC Scopus subject areas

Cite this

Investigating 2d and 3d convolutions for multitemporal land cover classification using remote sensing images. / Voelsen, M.; Teimouri, M.; Rottensteiner, F. et al.
In: ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol. 5, No. 3, 17.05.2022, p. 271-279.

Research output: Contribution to journalConference articleResearchpeer review

Voelsen, M, Teimouri, M, Rottensteiner, F & Heipke, C 2022, 'Investigating 2d and 3d convolutions for multitemporal land cover classification using remote sensing images', ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, vol. 5, no. 3, pp. 271-279. https://doi.org/10.5194/isprs-Annals-V-3-2022-271-2022
Voelsen, M., Teimouri, M., Rottensteiner, F., & Heipke, C. (2022). Investigating 2d and 3d convolutions for multitemporal land cover classification using remote sensing images. ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 5(3), 271-279. https://doi.org/10.5194/isprs-Annals-V-3-2022-271-2022
Voelsen M, Teimouri M, Rottensteiner F, Heipke C. Investigating 2d and 3d convolutions for multitemporal land cover classification using remote sensing images. ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences. 2022 May 17;5(3):271-279. doi: 10.5194/isprs-Annals-V-3-2022-271-2022
Voelsen, M. ; Teimouri, M. ; Rottensteiner, F. et al. / Investigating 2d and 3d convolutions for multitemporal land cover classification using remote sensing images. In: ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences. 2022 ; Vol. 5, No. 3. pp. 271-279.
Download
@article{7c676a0d0f9b435a9bd421706e30bbce,
title = "Investigating 2d and 3d convolutions for multitemporal land cover classification using remote sensing images",
abstract = "With the availability of large amounts of satellite image time series (SITS), the identification of different materials of the Earth's surface is possible with a high temporal resolution. One of the basic tasks is the pixel-wise classification of land cover, i.e.The task of identifying the physical material of the Earth's surface in an image. Fully convolutional neural networks (FCN) are successfully used for this task. In this paper, we investigate different FCN variants, using different methods for the computation of spatial, spectral, and temporal features. We investigate the impact of 3D convolutions in the spatial-Temporal as well as in the spatial-spectral dimensions in comparison to 2D convolutions in the spatial dimensions only. Additionally, we introduce a new method to generate multitemporal input patches by using time intervals instead of fixed acquisition dates. We then choose the image that is closest in time to the middle of the corresponding time interval, which makes our approach more flexible with respect to the requirements for the acquisition of new data. Using these multi-Temporal input patches, generated from Sentinel-2 images, we improve the classification of land cover by 4% in the mean F1-score and 1.3% in the overall accuracy compared to a classification using mono-Temporal input patches. Furthermore, the usage of 3D convolutions instead of 2D convolutions improves the classification performance by a small amount of 0.4% in the mean F1-score and 1.2% in the overall accuracy. ",
keywords = "3D-CNN, FCN, land cover classification, Multi-Temporal images, Remote sensing",
author = "M. Voelsen and M. Teimouri and F. Rottensteiner and C. Heipke",
year = "2022",
month = may,
day = "17",
doi = "10.5194/isprs-Annals-V-3-2022-271-2022",
language = "English",
volume = "5",
pages = "271--279",
number = "3",
note = "2022 24th ISPRS Congress on Imaging Today, Foreseeing Tomorrow, Commission III ; Conference date: 06-06-2022 Through 11-06-2022",

}

Download

TY - JOUR

T1 - Investigating 2d and 3d convolutions for multitemporal land cover classification using remote sensing images

AU - Voelsen, M.

AU - Teimouri, M.

AU - Rottensteiner, F.

AU - Heipke, C.

PY - 2022/5/17

Y1 - 2022/5/17

N2 - With the availability of large amounts of satellite image time series (SITS), the identification of different materials of the Earth's surface is possible with a high temporal resolution. One of the basic tasks is the pixel-wise classification of land cover, i.e.The task of identifying the physical material of the Earth's surface in an image. Fully convolutional neural networks (FCN) are successfully used for this task. In this paper, we investigate different FCN variants, using different methods for the computation of spatial, spectral, and temporal features. We investigate the impact of 3D convolutions in the spatial-Temporal as well as in the spatial-spectral dimensions in comparison to 2D convolutions in the spatial dimensions only. Additionally, we introduce a new method to generate multitemporal input patches by using time intervals instead of fixed acquisition dates. We then choose the image that is closest in time to the middle of the corresponding time interval, which makes our approach more flexible with respect to the requirements for the acquisition of new data. Using these multi-Temporal input patches, generated from Sentinel-2 images, we improve the classification of land cover by 4% in the mean F1-score and 1.3% in the overall accuracy compared to a classification using mono-Temporal input patches. Furthermore, the usage of 3D convolutions instead of 2D convolutions improves the classification performance by a small amount of 0.4% in the mean F1-score and 1.2% in the overall accuracy.

AB - With the availability of large amounts of satellite image time series (SITS), the identification of different materials of the Earth's surface is possible with a high temporal resolution. One of the basic tasks is the pixel-wise classification of land cover, i.e.The task of identifying the physical material of the Earth's surface in an image. Fully convolutional neural networks (FCN) are successfully used for this task. In this paper, we investigate different FCN variants, using different methods for the computation of spatial, spectral, and temporal features. We investigate the impact of 3D convolutions in the spatial-Temporal as well as in the spatial-spectral dimensions in comparison to 2D convolutions in the spatial dimensions only. Additionally, we introduce a new method to generate multitemporal input patches by using time intervals instead of fixed acquisition dates. We then choose the image that is closest in time to the middle of the corresponding time interval, which makes our approach more flexible with respect to the requirements for the acquisition of new data. Using these multi-Temporal input patches, generated from Sentinel-2 images, we improve the classification of land cover by 4% in the mean F1-score and 1.3% in the overall accuracy compared to a classification using mono-Temporal input patches. Furthermore, the usage of 3D convolutions instead of 2D convolutions improves the classification performance by a small amount of 0.4% in the mean F1-score and 1.2% in the overall accuracy.

KW - 3D-CNN

KW - FCN

KW - land cover classification

KW - Multi-Temporal images

KW - Remote sensing

UR - http://www.scopus.com/inward/record.url?scp=85132043839&partnerID=8YFLogxK

U2 - 10.5194/isprs-Annals-V-3-2022-271-2022

DO - 10.5194/isprs-Annals-V-3-2022-271-2022

M3 - Conference article

AN - SCOPUS:85132043839

VL - 5

SP - 271

EP - 279

JO - ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences

JF - ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences

SN - 2194-9042

IS - 3

T2 - 2022 24th ISPRS Congress on Imaging Today, Foreseeing Tomorrow, Commission III

Y2 - 6 June 2022 through 11 June 2022

ER -