elib
DLR-Header
DLR-Logo -> http://www.dlr.de
DLR Portal Home | Imprint | Privacy Policy | Contact | Deutsch
Fontsize: [-] Text [+]

Uncertainty is not sufficient for identifying noisy labels in training data for binary segmentation of building footprints

Ulman, Hannah and Gütter, Jonas Aaron and Niebling, Julia (2023) Uncertainty is not sufficient for identifying noisy labels in training data for binary segmentation of building footprints. Frontiers in Remote Sensing (3), p. 1100012. Frontiers Media S.A.. doi: 10.3389/frsen.2022.1100012. ISSN 2673-6187.

[img] PDF - Published version
1MB

Official URL: https://www.frontiersin.org/articles/10.3389/frsen.2022.1100012/full

Abstract

Obtaining high quality labels is a major challenge for the application of deep neural networks in the remote sensing domain. A common way of acquiring labels is the usage of crowd sourcing which can provide much needed training data sets but also often contains incorrect labels which can affect the training process of a deep neural network significantly. In this paper, we exploit uncertainty to identify a certain type of label noise for semantic segmentation of buildings in satellite imagery. That type of label noise is known as “omission noise,” i.e., missing labels for whole buildings which still appear in the satellite image. Following the literature, uncertainty during training can help in identifying the “sweet spot” between generalizing well and overfitting to label noise, which is further used to differentiate between noisy and clean labels. The differentiation between clean and noisy labels is based on pixel-wise uncertainty estimation and beta distribution fitting to the uncertainty estimates. For our study, we create a data set for building segmentation with different levels of omission noise to evaluate the impact of the noise level on the performance of the deep neural network during training. In doing so, we show that established uncertainty-based methods to identify noisy labels are in general not sufficient enough for our kind of remote sensing data. On the other hand, for some noise levels, we observe some promising differences between noisy and clean data which opens the possibility to refine the state-of-the-art methods further.

Item URL in elib:https://elib.dlr.de/193825/
Document Type:Article
Title:Uncertainty is not sufficient for identifying noisy labels in training data for binary segmentation of building footprints
Authors:
AuthorsInstitution or Email of AuthorsAuthor's ORCID iDORCID Put Code
Ulman, HannahUNSPECIFIEDUNSPECIFIEDUNSPECIFIED
Gütter, Jonas AaronUNSPECIFIEDUNSPECIFIEDUNSPECIFIED
Niebling, JuliaUNSPECIFIEDhttps://orcid.org/0000-0001-5413-2234UNSPECIFIED
Date:10 January 2023
Journal or Publication Title:Frontiers in Remote Sensing
Refereed publication:Yes
Open Access:Yes
Gold Open Access:Yes
In SCOPUS:No
In ISI Web of Science:Yes
DOI:10.3389/frsen.2022.1100012
Page Range:p. 1100012
Publisher:Frontiers Media S.A.
Series Name:Image Analysis and Classification
ISSN:2673-6187
Status:Published
Keywords:deep learning, remote sensing, uncertainty, label noise, segmentation
HGF - Research field:Aeronautics, Space and Transport
HGF - Program:Space
HGF - Program Themes:Space System Technology
DLR - Research area:Raumfahrt
DLR - Program:R SY - Space System Technology
DLR - Research theme (Project):R - Basic research in the field of machine learning
Location: Jena
Institutes and Institutions:Institute of Data Science > Data Analysis and Intelligence
Deposited By: Niebling, Julia
Deposited On:14 Feb 2023 14:06
Last Modified:28 Feb 2024 08:24

Repository Staff Only: item control page

Browse
Search
Help & Contact
Information
electronic library is running on EPrints 3.3.12
Website and database design: Copyright © German Aerospace Center (DLR). All rights reserved.