서브메뉴
검색
Quantitative and Efficient Scanning Transmission Electron Microscopy with Machine Learning
Quantitative and Efficient Scanning Transmission Electron Microscopy with Machine Learning
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20250211153129
- ISBN
- 9798346874058
- DDC
- 620.11
- 저자명
- Wei, Jingrui.
- 서명/저자
- Quantitative and Efficient Scanning Transmission Electron Microscopy with Machine Learning
- 발행사항
- [Sl] : The University of Wisconsin - Madison, 2024
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2024
- 형태사항
- 94 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 86-06, Section: B.
- 주기사항
- Advisor: Voyles, Paul M.
- 학위논문주기
- Thesis (Ph.D.)--The University of Wisconsin - Madison, 2024.
- 초록/해제
- 요약Scanning transmission electron microscopy (STEM) has advanced the field of materials science by enabling atomic-scale structural and functional imaging. Beyond standard 2D imaging, higher dimensional data acquisition and analysis are enabling critical new sample insights. New challenges arise as well from the fast data streaming and big data volume. This thesis will discuss how advanced computational approaches and machine learning (ML) enhance the efficiency of STEM data interpretation with example topics. First, we demonstrate convolutional neural network (CNN) approaches for precise atom localization in high-resolution STEM images, establishing comprehensive benchmarks and investigating the relationship between model characteristics and performance across varying image qualities and content. Second, we introduce a deep learning-based electron counting method for ultrafast 4D-STEM detectors, utilizing a Faster-RCNN architecture to achieve accurate electron event detection at higher electron dose than previous methods. Third, a physics-informed regression model for STEM aberration measurement was developed to provide rapid aberration estimation for fine probe alignment or a starting point for sample phase reconstruction algorithms. Fourth, Z-contrast STEM imaging and 4D-STEM were used to investigate short-range order in high entropy carbides with different compositions and thermal treatment. The thesis concludes with a chapter speculating on future directions for research using ML in electron microscopy. Throughout this work, we highlight the importance of incorporating prior knowledge into the machine learning framework, ranging from data representation to model design, contributing to the broader field of quantitative electron microscopy and automation of STEM analysis.
- 일반주제명
- Materials science
- 일반주제명
- Applied physics
- 일반주제명
- Computational physics
- 키워드
- Machine learning
- 키워드
- Electron dose
- 기타저자
- The University of Wisconsin - Madison Materials Science and Engineering
- 기본자료저록
- Dissertations Abstracts International. 86-06B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008250123s2024 us c eng d■001000017165150
■00520250211153129
■006m o d
■007cr#unu||||||||
■020 ▼a9798346874058
■035 ▼a(MiAaPQ)AAI31767615
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a620.11
■1001 ▼aWei, Jingrui.
■24510▼aQuantitative and Efficient Scanning Transmission Electron Microscopy with Machine Learning
■260 ▼a[Sl]▼bThe University of Wisconsin - Madison▼c2024
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2024
■300 ▼a94 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 86-06, Section: B.
■500 ▼aAdvisor: Voyles, Paul M.
■5021 ▼aThesis (Ph.D.)--The University of Wisconsin - Madison, 2024.
■520 ▼aScanning transmission electron microscopy (STEM) has advanced the field of materials science by enabling atomic-scale structural and functional imaging. Beyond standard 2D imaging, higher dimensional data acquisition and analysis are enabling critical new sample insights. New challenges arise as well from the fast data streaming and big data volume. This thesis will discuss how advanced computational approaches and machine learning (ML) enhance the efficiency of STEM data interpretation with example topics. First, we demonstrate convolutional neural network (CNN) approaches for precise atom localization in high-resolution STEM images, establishing comprehensive benchmarks and investigating the relationship between model characteristics and performance across varying image qualities and content. Second, we introduce a deep learning-based electron counting method for ultrafast 4D-STEM detectors, utilizing a Faster-RCNN architecture to achieve accurate electron event detection at higher electron dose than previous methods. Third, a physics-informed regression model for STEM aberration measurement was developed to provide rapid aberration estimation for fine probe alignment or a starting point for sample phase reconstruction algorithms. Fourth, Z-contrast STEM imaging and 4D-STEM were used to investigate short-range order in high entropy carbides with different compositions and thermal treatment. The thesis concludes with a chapter speculating on future directions for research using ML in electron microscopy. Throughout this work, we highlight the importance of incorporating prior knowledge into the machine learning framework, ranging from data representation to model design, contributing to the broader field of quantitative electron microscopy and automation of STEM analysis.
■590 ▼aSchool code: 0262.
■650 4▼aMaterials science
■650 4▼aApplied physics
■650 4▼aComputational physics
■653 ▼aScanning transmission electron microscopy
■653 ▼aElectron microscopy
■653 ▼aMachine learning
■653 ▼aStructure characterization
■653 ▼aElectron dose
■690 ▼a0794
■690 ▼a0800
■690 ▼a0215
■690 ▼a0216
■71020▼aThe University of Wisconsin - Madison▼bMaterials Science and Engineering.
■7730 ▼tDissertations Abstracts International▼g86-06B.
■790 ▼a0262
■791 ▼aPh.D.
■792 ▼a2024
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17165150▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.
