Self-supervised learning of deep despeckling networks with MERLIN: ensuring the statistical independence of the real and imaginary parts

Emanuele Dalsasso; Frédéric Brigui; Loïc Denis; Rémy Abergel; Florence Tupin

Self-supervised learning of deep despeckling networks with MERLIN: ensuring the statistical independence of the real and imaginary parts

Emanuele Dalsasso
, Frédéric Brigui
, Loïc Denis
, Rémy Abergel
, Florence Tupin

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

Abstract

Due to the wide variety of sensors, with different spatial resolutions, operating frequency bands, as well as acquisition modes (Stripmap, Spotlight, TOPS...), despeckling neural networks trained on a given type of SAR images do not generalize well. By directly training on images from the sensor and acquisition mode of interest, self-supervised learning is a very appealing solution. This paper analyses the preprocessing requirements of the MERLIN strategy that assumes statistical independence of the real and imaginary parts of single-look-complex SAR images to perform the self-supervised training. Adequate spectral corrections are proposed to handle asymmetrical spectra and moving Doppler centroids.

Original language	English
Title of host publication	EUSAR 2024 - 15th European Conference on Synthetic Aperture Radar
Publisher	Institute of Electrical and Electronics Engineers Inc.
Pages	254-259
Number of pages	6
ISBN (Electronic)	9783800762873
Publication status	Published - 1 Jan 2024
Event	15th European Conference on Synthetic Aperture Radar, EUSAR 2024 - Munich, Germany Duration: 23 Apr 2024 → 26 Apr 2024

Publication series

Name	Proceedings of the European Conference on Synthetic Aperture Radar, EUSAR
ISSN (Print)	2197-4403

Conference

Conference	15th European Conference on Synthetic Aperture Radar, EUSAR 2024
Country/Territory	Germany
City	Munich
Period	23/04/24 → 26/04/24

Cite this

Dalsasso, E., Brigui, F., Denis, L., Abergel, R., & Tupin, F. (2024). Self-supervised learning of deep despeckling networks with MERLIN: ensuring the statistical independence of the real and imaginary parts. In EUSAR 2024 - 15th European Conference on Synthetic Aperture Radar (pp. 254-259). (Proceedings of the European Conference on Synthetic Aperture Radar, EUSAR). Institute of Electrical and Electronics Engineers Inc..

Dalsasso, Emanuele ; Brigui, Frédéric ; Denis, Loïc et al. / Self-supervised learning of deep despeckling networks with MERLIN : ensuring the statistical independence of the real and imaginary parts. EUSAR 2024 - 15th European Conference on Synthetic Aperture Radar. Institute of Electrical and Electronics Engineers Inc., 2024. pp. 254-259 (Proceedings of the European Conference on Synthetic Aperture Radar, EUSAR).

@inproceedings{1343955278144e89bcfb47d1b111271d,

title = "Self-supervised learning of deep despeckling networks with MERLIN: ensuring the statistical independence of the real and imaginary parts",

abstract = "Due to the wide variety of sensors, with different spatial resolutions, operating frequency bands, as well as acquisition modes (Stripmap, Spotlight, TOPS...), despeckling neural networks trained on a given type of SAR images do not generalize well. By directly training on images from the sensor and acquisition mode of interest, self-supervised learning is a very appealing solution. This paper analyses the preprocessing requirements of the MERLIN strategy that assumes statistical independence of the real and imaginary parts of single-look-complex SAR images to perform the self-supervised training. Adequate spectral corrections are proposed to handle asymmetrical spectra and moving Doppler centroids.",

author = "Emanuele Dalsasso and Fr{\'e}d{\'e}ric Brigui and Lo{\"i}c Denis and R{\'e}my Abergel and Florence Tupin",

note = "Publisher Copyright: {\textcopyright} VDE VERLAG GMBH ∙ Berlin ∙ Offenbach.; 15th European Conference on Synthetic Aperture Radar, EUSAR 2024 ; Conference date: 23-04-2024 Through 26-04-2024",

year = "2024",

month = jan,

day = "1",

language = "English",

series = "Proceedings of the European Conference on Synthetic Aperture Radar, EUSAR",

publisher = "Institute of Electrical and Electronics Engineers Inc.",

pages = "254--259",

booktitle = "EUSAR 2024 - 15th European Conference on Synthetic Aperture Radar",

}

Dalsasso, E, Brigui, F, Denis, L, Abergel, R & Tupin, F 2024, Self-supervised learning of deep despeckling networks with MERLIN: ensuring the statistical independence of the real and imaginary parts. in EUSAR 2024 - 15th European Conference on Synthetic Aperture Radar. Proceedings of the European Conference on Synthetic Aperture Radar, EUSAR, Institute of Electrical and Electronics Engineers Inc., pp. 254-259, 15th European Conference on Synthetic Aperture Radar, EUSAR 2024, Munich, Germany, 23/04/24.

Self-supervised learning of deep despeckling networks with MERLIN: ensuring the statistical independence of the real and imaginary parts. / Dalsasso, Emanuele; Brigui, Frédéric; Denis, Loïc et al.
EUSAR 2024 - 15th European Conference on Synthetic Aperture Radar. Institute of Electrical and Electronics Engineers Inc., 2024. p. 254-259 (Proceedings of the European Conference on Synthetic Aperture Radar, EUSAR).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Self-supervised learning of deep despeckling networks with MERLIN

T2 - 15th European Conference on Synthetic Aperture Radar, EUSAR 2024

AU - Dalsasso, Emanuele

AU - Brigui, Frédéric

AU - Denis, Loïc

AU - Abergel, Rémy

AU - Tupin, Florence

PY - 2024/1/1

Y1 - 2024/1/1

N2 - Due to the wide variety of sensors, with different spatial resolutions, operating frequency bands, as well as acquisition modes (Stripmap, Spotlight, TOPS...), despeckling neural networks trained on a given type of SAR images do not generalize well. By directly training on images from the sensor and acquisition mode of interest, self-supervised learning is a very appealing solution. This paper analyses the preprocessing requirements of the MERLIN strategy that assumes statistical independence of the real and imaginary parts of single-look-complex SAR images to perform the self-supervised training. Adequate spectral corrections are proposed to handle asymmetrical spectra and moving Doppler centroids.

AB - Due to the wide variety of sensors, with different spatial resolutions, operating frequency bands, as well as acquisition modes (Stripmap, Spotlight, TOPS...), despeckling neural networks trained on a given type of SAR images do not generalize well. By directly training on images from the sensor and acquisition mode of interest, self-supervised learning is a very appealing solution. This paper analyses the preprocessing requirements of the MERLIN strategy that assumes statistical independence of the real and imaginary parts of single-look-complex SAR images to perform the self-supervised training. Adequate spectral corrections are proposed to handle asymmetrical spectra and moving Doppler centroids.

M3 - Conference contribution

AN - SCOPUS:85193951844

T3 - Proceedings of the European Conference on Synthetic Aperture Radar, EUSAR

SP - 254

EP - 259

BT - EUSAR 2024 - 15th European Conference on Synthetic Aperture Radar

PB - Institute of Electrical and Electronics Engineers Inc.

Y2 - 23 April 2024 through 26 April 2024

ER -

Dalsasso E, Brigui F, Denis L, Abergel R, Tupin F. Self-supervised learning of deep despeckling networks with MERLIN: ensuring the statistical independence of the real and imaginary parts. In EUSAR 2024 - 15th European Conference on Synthetic Aperture Radar. Institute of Electrical and Electronics Engineers Inc. 2024. p. 254-259. (Proceedings of the European Conference on Synthetic Aperture Radar, EUSAR).

Self-supervised learning of deep despeckling networks with MERLIN: ensuring the statistical independence of the real and imaginary parts

Abstract

Publication series

Conference

Fingerprint

Cite this