Please use this identifier to cite or link to this item:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/90740
Title: | 基於深度學習的維納反捲積方法應用於彩色共焦顯微術之光學解析度提升 Enhancing Optical Resolution in Chromatic Confocal Microscopy: A Deep Learning-Based Wiener Deconvolution Approach |
Authors: | 盛杉奕 Sanjeev Kumar Singh |
Advisor: | 陳亮嘉 Liang-Chia Chen |
Keyword: | 點掃描色差共焦顯微術(PSCCM),深度學習維納解卷積,基於波理論的標量繞射參數模型,實驗評估,微觀特徵,解析度改善, Point Scanning Chromatic Confocal Microscopy (PSCCM),Deep Learning Weiner Deconvolution,Wave Theory-based Scalar Diffraction Parametric Model,Experimental Evaluations,Microscopic Features,Resolution Improvements, |
Publication Year : | 2023 |
Degree: | 碩士 |
Abstract: | 此論文旨於通過創新技術來提升點掃描色散共焦顯微鏡(PSCCM)的成像能力。主要目標是使用深度學習的維納去卷積和基於波動理論的標量繞射參數模型,來增強橫向和軸向的解析度。
為了克服PSCCM的解析度限制,我們應用了深度學習的維納去卷積。使用深度學習開發了一種強大的去卷積算法,有效地提高了成像質量並增強了橫向和軸向的解析度。 此外,引入了一個標量繞射參數模型以進一步增強PSCCM的解析度。該模型考慮了由光學元件引起的繞射效應,並利用優化技術來優化成像系統。所提出的模型在橫向和軸向的解析度方面都顯示出了顯著的改進,推動了PSCCM性能的極限。 透過使用合成和真實的樣品進行廣泛的實驗評估,驗證了所提出技術的有效性。成功地測量和描述了各種具有挑戰性的樣品結構,例如1微米的階高、公稱尺寸為W8/S8的淺型光柵結構以及25微米的凸塊。所獲得的結果展示了該方法對於準確分析這些複雜結構的能力,顯著地提高了成像解析度,並使得微觀特徵的詳細可視化和分析成為可能。經實驗結果驗證,透過採用超過50微米閥值的點擴散函數(PSFs),深度學習的去卷積過程將更有能力處理更廣泛的結構範疇,包括那些小於50微米範圍的三維結構。這種策略性的擴展無疑將有助於量測的準確性,提供了一個更全面和綜合的解決方案,用於尺寸分析。 通過整合深度學習的維納去卷積和一個標量繞射參數模型,此論文為PSCCM領域做出了貢獻,並為生物醫學成像、材料科學和納米技術領域的應用打開了新的可能性。所提出的技術為解析度的大幅改進提供了巨大的潛力,增強了PSCCM在廣泛的研究和工業應用中的潛力。 This thesis focuses on advancing the imaging capabilities of point scanning chromatic confocal microscopy (PSCCM) through innovative techniques. The main goal is to enhance both lateral and axial resolution using deep learning Weiner deconvolution and a wave theory-based scalar diffraction parametric model. To overcome the resolution limitations of PSCCM, deep learning Weiner deconvolution is applied. A robust deconvolution algorithm is developed using deep learning, effectively improving the imaging quality and enhance lateral and axial resolution. In addition, a scalar diffraction parametric model is introduced to further enhance PSCCM's resolution. This model accounts for diffraction effects caused by optical elements and utilizes optimization techniques to optimize the imaging system. The proposed model demonstrates remarkable improvements in both lateral and axial resolution, pushing the boundaries of PSCCM performance. The effectiveness of the proposed techniques is validated through extensive experimental evaluations using synthetic and real-world samples. Various challenging sample structures, such as a 1 µm step height, a shallow grating structure with a nominal dimension of W8/S8, and a 25 µm bump, are successfully measured and characterized. The obtained results showcase the method's capability to accurately analyze these complex structures, significantly advancing imaging resolution and enabling detailed visualization and analysis of microscopic features. By incorporating PSFs for heights beyond the 50 μm threshold, the deep learning deconvolution process would be better equipped to tackle a broader spectrum of structures, including those below the 50 μm range. This strategic expansion would undoubtedly contribute to the refinement and accuracy of the measurement process, offering a more holistic and comprehensive solution for dimensional analysis. By integrating deep learning Weiner deconvolution and a scalar diffraction parametric model, this thesis contributes to the field of PSCCM and opens up new possibilities for applications in biomedical imaging, material science, and nanotechnology. The proposed techniques offer substantial resolution improvements, enhancing PSCCM's potential for a wide range of research and industrial applications. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/90740 |
DOI: | 10.6342/NTU202304007 |
Fulltext Rights: | 未授權 |
Appears in Collections: | 機械工程學系 |
Files in This Item:
File | Size | Format | |
---|---|---|---|
ntu-111-2.pdf Restricted Access | 8.37 MB | Adobe PDF |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.