Learning a versatile representation of SAR data for regression and segmentation by leveraging self-supervised despeckling with MERLIN

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Synthetic Aperture Radar (SAR) images are abundantly available, yet labels are often missing. Thus, training a neural network in a fully supervised manner is arduous. In this work, we leverage MERLIN, a self-supervised despeckling algorithm, to learn a mapping of SAR images into a representation space shared among despeckling, building segmentation and height regression. Our experiments demonstrate that the joint training of a neural network for these three tasks reduces considerably the need for labeled data to solve the supervised tasks: positive results are obtained even when only 1% of the dataset is annotated.

Original languageEnglish
Title of host publicationEUSAR 2024 - 15th European Conference on Synthetic Aperture Radar
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages1265-1270
Number of pages6
ISBN (Electronic)9783800762873
Publication statusPublished - 1 Jan 2024
Event15th European Conference on Synthetic Aperture Radar, EUSAR 2024 - Munich, Germany
Duration: 23 Apr 202426 Apr 2024

Publication series

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

Conference

Conference15th European Conference on Synthetic Aperture Radar, EUSAR 2024
Country/TerritoryGermany
CityMunich
Period23/04/2426/04/24

Fingerprint

Dive into the research topics of 'Learning a versatile representation of SAR data for regression and segmentation by leveraging self-supervised despeckling with MERLIN'. Together they form a unique fingerprint.

Cite this