Deep Learning of Crystalline Defects from TEM images: A Solution for the Problem of" Never Enough Training Data"

Crystalline defects, such as line-like dislocations, play an important role for the performance and reliability of many metallic devices. Their interaction and evolution still poses a multitude of open questions to materials science and materials physics. In-situ TEM experiments can provide importan...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:Machine learning: science and technology
Hlavní autoři: Govind, Kishan, Oliveros, Daniela, Dlouhy, Antonin, Legros, Marc, Sandfeld, Stefan
Médium: Journal Article
Jazyk:angličtina
Vydáno: IOP Publishing Ltd 12.07.2023
Témata:
ISSN:2632-2153
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Abstract Crystalline defects, such as line-like dislocations, play an important role for the performance and reliability of many metallic devices. Their interaction and evolution still poses a multitude of open questions to materials science and materials physics. In-situ TEM experiments can provide important insights into how dislocations behave and move. During such experiments, the dislocation microstructure is captured in form of videos. The analysis of individual video frames can provide useful insights but is limited by the capabilities of automated identification, digitization, and quantitative extraction of the dislocations as curved objects. The vast amount of data also makes manual annotation very time consuming, thereby limiting the use of Deep Learning-based, automated image analysis and segmentation of the dislocation microstructure. In this work, a parametric model for generating synthetic training data for segmentation of dislocations is developed. Even though domain scientists might dismiss synthetic training images sometimes as too artificial, our findings show that they can result in superior performance, particularly regarding the generalizing of the Deep Learning models with respect to different microstructures and imaging conditions. Additionally, we propose an enhanced deep learning method optimized for segmenting overlapping or intersecting dislocation lines. Upon testing this framework on four distinct real datasets, we find that our synthetic training data are able to yield high-quality results also on real images-even more so if fine-tune on a few real images was done. Our approach demonstrates the potential of synthetic data in overcoming the limitations of manual annotation in TEM, paving the way for more efficient and accurate analysis of dislocation microstructures. Last but not least, segmenting such thin, curvilinear structures is a task that is ubiquitous in many fields, which makes our method a potential candidate for other applications as well.
AbstractList Crystalline defects, such as line-like dislocations, play an important role for the performance and reliability of many metallic devices. Their interaction and evolution still poses a multitude of open questions to materials science and materials physics. In-situ TEM experiments can provide important insights into how dislocations behave and move. During such experiments, the dislocation microstructure is captured in form of videos. The analysis of individual video frames can provide useful insights but is limited by the capabilities of automated identification, digitization, and quantitative extraction of the dislocations as curved objects. The vast amount of data also makes manual annotation very time consuming, thereby limiting the use of Deep Learning-based, automated image analysis and segmentation of the dislocation microstructure. In this work, a parametric model for generating synthetic training data for segmentation of dislocations is developed. Even though domain scientists might dismiss synthetic training images sometimes as too artificial, our findings show that they can result in superior performance, particularly regarding the generalizing of the Deep Learning models with respect to different microstructures and imaging conditions. Additionally, we propose an enhanced deep learning method optimized for segmenting overlapping or intersecting dislocation lines. Upon testing this framework on four distinct real datasets, we find that our synthetic training data are able to yield high-quality results also on real images-even more so if fine-tune on a few real images was done. Our approach demonstrates the potential of synthetic data in overcoming the limitations of manual annotation in TEM, paving the way for more efficient and accurate analysis of dislocation microstructures. Last but not least, segmenting such thin, curvilinear structures is a task that is ubiquitous in many fields, which makes our method a potential candidate for other applications as well.
Author Legros, Marc
Oliveros, Daniela
Dlouhy, Antonin
Sandfeld, Stefan
Govind, Kishan
Author_xml – sequence: 1
  givenname: Kishan
  surname: Govind
  fullname: Govind, Kishan
  organization: Forschungszentrum Jülich GmbH | Centre de recherche de Jülich | Juelich Research Centre
– sequence: 2
  givenname: Daniela
  surname: Oliveros
  fullname: Oliveros, Daniela
  organization: Centre d'élaboration de matériaux et d'études structurales
– sequence: 3
  givenname: Antonin
  surname: Dlouhy
  fullname: Dlouhy, Antonin
  organization: Czech Academy of Sciences [Prague]
– sequence: 4
  givenname: Marc
  orcidid: 0000-0001-5830-0434
  surname: Legros
  fullname: Legros, Marc
  organization: Centre d'élaboration de matériaux et d'études structurales
– sequence: 5
  givenname: Stefan
  orcidid: 0000-0001-9560-4728
  surname: Sandfeld
  fullname: Sandfeld, Stefan
  organization: Forschungszentrum Jülich GmbH | Centre de recherche de Jülich | Juelich Research Centre
BackLink https://hal.science/hal-04298752$$DView record in HAL
BookMark eNotjktPwkAURidGExH5Ae5u2LkozqPTad0RQDGpj8Qu3DW3cAfGlA6ZFhL-PfWxOl_O4uS7YZeNb4ixO8Encao1f8Dw5Y4TqbiZ8ERJecEGsmckhVbXbNS235xzqYXSkg_Yfk60h5wwNK7ZgLcwC6e2w7p2DcGcLK26FmzwOygWr-B2uKH2Eabw6etD53wD1gfotgQfwVc17frEGN7oSAEWjT9stlAEdL_xOXY4vmVXFuuWRv8csuJpUcyWUf7-_DKb5hHqVESVsUannMxaoBUmiSu1Vv2seqm0tSKpDGZEwlJWpanEeFUllFkjVolGu1ZDdv-X3WJd7kP_O5xKj65cTvPyx_FYZqnR8ijUGU1eXqk
ContentType Journal Article
Copyright Distributed under a Creative Commons Attribution 4.0 International License
Copyright_xml – notice: Distributed under a Creative Commons Attribution 4.0 International License
DBID 1XC
VOOES
DOI 10.48550/arXiv.2307.06322
DatabaseName Hyper Article en Ligne (HAL)
Hyper Article en Ligne (HAL) (Open Access)
DatabaseTitleList
DeliveryMethod fulltext_linktorsrc
Discipline Physics
Computer Science
EISSN 2632-2153
ExternalDocumentID oai:HAL:hal-04298752v1
GroupedDBID 1XC
88I
ABHWH
ABUWG
ACHIP
AEINN
AFFHD
AFKRA
AKPSB
ALMA_UNASSIGNED_HOLDINGS
ARAPS
AZQEC
BENPR
BGLVJ
CCPQU
CJUJL
DWQXO
EBS
GNUQQ
GROUPED_DOAJ
HCIFZ
IOP
K7-
M2P
M~E
N5L
O3W
OK1
PHGZM
PHGZT
PIMPY
PQGLB
VOOES
ID FETCH-LOGICAL-a581-b7f7580e7d1af1764b3d31afb58035ff16b7a9ee1fe9b882a4cb6e9f71c65afd3
IngestDate Sat Nov 01 14:59:50 EDT 2025
IsDoiOpenAccess true
IsOpenAccess true
IsPeerReviewed true
IsScholarly true
Keywords dislocation
deep learning
Computer Vision and Pattern Recognition (cs.CV)
FOS: Computer and information sciences
data mining
deep learning synthetic training data segmentation data mining transmission electron microscopy dislocation crystal defect
segmentation
FOS: Physical sciences
Materials Science (cond-mat.mtrl-sci)
crystal defect
transmission electron microscopy
synthetic training data
Language English
License Distributed under a Creative Commons Attribution 4.0 International License: http://creativecommons.org/licenses/by/4.0
LinkModel OpenURL
MergedId FETCHMERGED-LOGICAL-a581-b7f7580e7d1af1764b3d31afb58035ff16b7a9ee1fe9b882a4cb6e9f71c65afd3
ORCID 0000-0001-9560-4728
0000-0001-5830-0434
OpenAccessLink http://dx.doi.org/10.48550/arXiv.2307.06322
ParticipantIDs hal_primary_oai_HAL_hal_04298752v1
PublicationCentury 2000
PublicationDate 2023-07-12
PublicationDateYYYYMMDD 2023-07-12
PublicationDate_xml – month: 07
  year: 2023
  text: 2023-07-12
  day: 12
PublicationDecade 2020
PublicationTitle Machine learning: science and technology
PublicationYear 2023
Publisher IOP Publishing Ltd
Publisher_xml – name: IOP Publishing Ltd
SSID ssj0002513520
Score 2.225091
SecondaryResourceType preprint
Snippet Crystalline defects, such as line-like dislocations, play an important role for the performance and reliability of many metallic devices. Their interaction and...
SourceID hal
SourceType Open Access Repository
SubjectTerms Computer Science
Physics
Title Deep Learning of Crystalline Defects from TEM images: A Solution for the Problem of" Never Enough Training Data"
URI https://hal.science/hal-04298752
hasFullText 1
inHoldings 1
isFullTextHit
isPrint
journalDatabaseRights – providerCode: PRVAON
  databaseName: DOAJ Directory of Open Access Journals
  databaseCode: DOA
  dateStart: 20200101
  customDbUrl:
  isFulltext: true
  eissn: 2632-2153
  dateEnd: 99991231
  titleUrlDefault: https://www.doaj.org/
  omitProxy: false
  ssIdentifier: ssj0002513520
  providerName: Directory of Open Access Journals
– providerCode: PRVIOP
  databaseName: Institute of Physics Journals Open Access
  databaseCode: O3W
  dateStart: 20200301
  customDbUrl:
  isFulltext: true
  eissn: 2632-2153
  dateEnd: 99991231
  titleUrlDefault: http://iopscience.iop.org/
  omitProxy: false
  ssIdentifier: ssj0002513520
  providerName: IOP Publishing
– providerCode: PRVHPJ
  databaseName: ROAD: Directory of Open Access Scholarly Resources
  databaseCode: M~E
  dateStart: 20200101
  customDbUrl:
  isFulltext: true
  eissn: 2632-2153
  dateEnd: 99991231
  titleUrlDefault: https://road.issn.org
  omitProxy: false
  ssIdentifier: ssj0002513520
  providerName: ISSN International Centre
– providerCode: PRVPQU
  databaseName: Computer Science Database
  databaseCode: K7-
  dateStart: 20200301
  customDbUrl:
  isFulltext: true
  eissn: 2632-2153
  dateEnd: 99991231
  titleUrlDefault: http://search.proquest.com/compscijour
  omitProxy: false
  ssIdentifier: ssj0002513520
  providerName: ProQuest
– providerCode: PRVPQU
  databaseName: ProQuest Central
  databaseCode: BENPR
  dateStart: 20200301
  customDbUrl:
  isFulltext: true
  eissn: 2632-2153
  dateEnd: 99991231
  titleUrlDefault: https://www.proquest.com/central
  omitProxy: false
  ssIdentifier: ssj0002513520
  providerName: ProQuest
– providerCode: PRVPQU
  databaseName: Publicly Available Content Database
  databaseCode: PIMPY
  dateStart: 20200301
  customDbUrl:
  isFulltext: true
  eissn: 2632-2153
  dateEnd: 99991231
  titleUrlDefault: http://search.proquest.com/publiccontent
  omitProxy: false
  ssIdentifier: ssj0002513520
  providerName: ProQuest
– providerCode: PRVPQU
  databaseName: Science Database
  databaseCode: M2P
  dateStart: 20200301
  customDbUrl:
  isFulltext: true
  eissn: 2632-2153
  dateEnd: 99991231
  titleUrlDefault: https://search.proquest.com/sciencejournals
  omitProxy: false
  ssIdentifier: ssj0002513520
  providerName: ProQuest
link http://cvtisr.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwtV1db9MwFLWqARIvfAwQn5NVwRMKNB-OG96idaMTa1eJPPQtclJ7qxTSqt2q8cIv4kdybMdtBjyMB16syG2sKPfUPvf23HsJeSviImaJ4F7Jg9DTugsPrMj3BGN93OGHvFSm2QQfj_vTaTLpdH66XJhNxeu6f32dLP-rqTEHY-vU2X8w93ZRTOAaRscIs2O8leEHUi5d2VQjaD5cfQcFrAyfHEgr3zBZJdnR6P38GzaUtc1PdyGyrfRwYrvNYJF3QYD9cKMVIbXp65M1rSUAG80_gzbJHRmBpnQdKUxCtUsfMoLNP8L5n3Vcw0ZddTfoLWDPKi0bsUpAmwy_PUUG1eLKIiTVbZDnLWnReXPHCJZtRzUAE12rcucDn5xN2nE4p3CSZmPUJeY9UJXwb0eAqdCmD7jVdL4xMvcPYGHNa7hRbnuYfs0ng-P89GT85eanLY3iMD3FeCEqT5_ccO6CDXztOwFnid5DRz92ET0wRbDZ5q9z8xgff38IEJgLF7A3BCZ7RB40ngdNLWIek46s98lD19WDNpv8PrlnRMHl-glZaihRByW6ULQFJdpAiWooUUCJWih9oil1QKIAEgWQaAMkLNGlBkbUwog6GFENo-5Tkh0fZYdDr-nQgd9y3_cKruBu9iSf-UL5PI6KcBbissBkyJTy44KLREpfyaSAKyeisohlorhfxkyoWfiM7NWLWj4nNFZlICRjKkpmUVyKgmHviHoCq6geK8IXpIs3ly9tCZZcF0WHaXI9tzPMy9t86RW5vwPca7J3ubqSb8jdcnM5X68OTEzmwBj2F25Ef-E
linkProvider ISSN International Centre
openUrl ctx_ver=Z39.88-2004&ctx_enc=info%3Aofi%2Fenc%3AUTF-8&rfr_id=info%3Asid%2Fsummon.serialssolutions.com&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.atitle=Deep+Learning+of+Crystalline+Defects+from+TEM+images%3A+A+Solution+for+the+Problem+of%22+Never+Enough+Training+Data%22&rft.jtitle=Machine+learning%3A+science+and+technology&rft.au=Govind%2C+Kishan&rft.au=Oliveros%2C+Daniela&rft.au=Dlouhy%2C+Antonin&rft.au=Legros%2C+Marc&rft.date=2023-07-12&rft.pub=IOP+Publishing+Ltd&rft.eissn=2632-2153&rft_id=info:doi/10.48550%2FarXiv.2307.06322&rft.externalDBID=HAS_PDF_LINK&rft.externalDocID=oai%3AHAL%3Ahal-04298752v1