서브메뉴
검색
Data-Driven Design of High-Dimensional, Snapshot Computational Imaging Systems
Data-Driven Design of High-Dimensional, Snapshot Computational Imaging Systems
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202105109
- ISBN
- 9798297601178
- DDC
- 535
- 저자명
- Markley, Eric.
- 서명/저자
- Data-Driven Design of High-Dimensional, Snapshot Computational Imaging Systems
- 발행사항
- [Sl] : University of California, Berkeley, 2025
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2025
- 형태사항
- 128 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-04, Section: B.
- 주기사항
- Advisor: Waller, Laura.
- 학위논문주기
- Thesis (Ph.D.)--University of California, Berkeley, 2025.
- 초록/해제
- 요약Modern imaging systems increasingly rely on computational methods to extract high-dimensional information from 2D optical measurements. Examples include snapshot 3D microscopy systems that capture volumetric data in a single exposure and hyperspectral imagers that simultaneously measure spatial and spectral information across dozens of wavelength channels. Designing such systems is challenging because it requires jointly optimizing both the optical hardware that encodes the scene and the computational algorithms that decode the measurements, a process complicated by the non-convex, high-dimensional parameter spaces and computationally expensive end-to-end training requirements. In this dissertation, we present data-driven approaches that address these challenges through physics-based simulation and information-theoretic design principles. We first develop a memory-efficient, end-to-end pipeline that jointly optimizes optical elements and neural reconstruction algorithms using differentiable simulation, demonstrating this method on a snapshot 3D fluorescence microscope that achieves improved resolution over heuristic designs. We then present a compact snapshot hyperspectral fluorescence microscope with a custom iterative reconstruction algorithm tailored to its physical model.To overcome the computational limitations of end-to-end optimization and accommodate non-differentiable reconstruction algorithms, we develop an information-theoretic optimization framework that treats optical design as a mutual information maximization problem. This approach, implemented through the IDEAL and IDEAL-IO methods, decouples encoder design from specific reconstruction implementations. By directly maximizing the information content of measurements rather than optimizing reconstruction fidelity, this framework provides a generalizable design principle that transcends particular decoder architectures while reducing the computational requirements in comparison to end-to-end design.The methods developed in this dissertation demonstrate that principled, simulation-driven design can achieve improved performance across diverse high-dimensional imaging modalities while maintaining computational tractability.
- 일반주제명
- Optics
- 일반주제명
- Computer science
- 일반주제명
- Bioinformatics
- 일반주제명
- Medical imaging
- 키워드
- Optimization
- 기타저자
- University of California, Berkeley Bioengineering
- 기본자료저록
- Dissertations Abstracts International. 87-04B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2025 us c eng d■001000017359369
■00520260202105109
■006m o d
■007cr#unu||||||||
■020 ▼a9798297601178
■035 ▼a(MiAaPQ)AAI32236909
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a535
■1001 ▼aMarkley, Eric.
■24510▼aData-Driven Design of High-Dimensional, Snapshot Computational Imaging Systems
■260 ▼a[Sl]▼bUniversity of California, Berkeley▼c2025
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2025
■300 ▼a128 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-04, Section: B.
■500 ▼aAdvisor: Waller, Laura.
■5021 ▼aThesis (Ph.D.)--University of California, Berkeley, 2025.
■520 ▼aModern imaging systems increasingly rely on computational methods to extract high-dimensional information from 2D optical measurements. Examples include snapshot 3D microscopy systems that capture volumetric data in a single exposure and hyperspectral imagers that simultaneously measure spatial and spectral information across dozens of wavelength channels. Designing such systems is challenging because it requires jointly optimizing both the optical hardware that encodes the scene and the computational algorithms that decode the measurements, a process complicated by the non-convex, high-dimensional parameter spaces and computationally expensive end-to-end training requirements. In this dissertation, we present data-driven approaches that address these challenges through physics-based simulation and information-theoretic design principles. We first develop a memory-efficient, end-to-end pipeline that jointly optimizes optical elements and neural reconstruction algorithms using differentiable simulation, demonstrating this method on a snapshot 3D fluorescence microscope that achieves improved resolution over heuristic designs. We then present a compact snapshot hyperspectral fluorescence microscope with a custom iterative reconstruction algorithm tailored to its physical model.To overcome the computational limitations of end-to-end optimization and accommodate non-differentiable reconstruction algorithms, we develop an information-theoretic optimization framework that treats optical design as a mutual information maximization problem. This approach, implemented through the IDEAL and IDEAL-IO methods, decouples encoder design from specific reconstruction implementations. By directly maximizing the information content of measurements rather than optimizing reconstruction fidelity, this framework provides a generalizable design principle that transcends particular decoder architectures while reducing the computational requirements in comparison to end-to-end design.The methods developed in this dissertation demonstrate that principled, simulation-driven design can achieve improved performance across diverse high-dimensional imaging modalities while maintaining computational tractability.
■590 ▼aSchool code: 0028.
■650 4▼aOptics
■650 4▼aComputer science
■650 4▼aBioinformatics
■650 4▼aMedical imaging
■653 ▼aData-driven design
■653 ▼aEnd-to-end design
■653 ▼aInformation theory
■653 ▼aOptimization
■653 ▼a3D fluorescence microscope
■690 ▼a0752
■690 ▼a0984
■690 ▼a0574
■690 ▼a0715
■71020▼aUniversity of California, Berkeley▼bBioengineering.
■7730 ▼tDissertations Abstracts International▼g87-04B.
■790 ▼a0028
■791 ▼aPh.D.
■792 ▼a2025
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17359369▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


