透過深度學習神經網路進行無透鏡數位全像顯微系統之影像重建與相位恢復;Image Reconstruction and Phase Recovery of Lensless Digital Holographic Microscopy System Through Deep Learning in Neural Network

NCU Institutional Repository > 生醫理工學院 > 生物醫學工程研究所 > 博碩士論文 > Item 987654321/94844

jsp.display-item.identifier=請使用永久網址來引用或連結此文件: http://ir.lib.ncu.edu.tw/handle/987654321/94844

题名:	透過深度學習神經網路進行無透鏡數位全像顯微系統之影像重建與相位恢復;Image Reconstruction and Phase Recovery of Lensless Digital Holographic Microscopy System Through Deep Learning in Neural Network
作者:	郭逸浦;Kuo, Yi-Pu
贡献者:	生醫科學與工程學系
关键词:	全像術;深度學習;影像還原;微型化系統;holography;deep learning;image reconstruction;image quality enhancement;miniaturized system
日期:	2024-07-24
上传时间:	2024-10-09 15:33:26 (UTC+8)
出版者:	國立中央大學
摘要:	光學複合顯微鏡一直是顯微成像的主要工具，其中視野範圍擴大、解析度提升和相位資訊之取得是光學成像中長期追求的重要目標，但傳統顯微系統其性能依賴於相對複雜、笨重且昂貴的鏡頭和對準機制等，本研究旨在透過深度學習神經網路進行無透鏡數位全像顯微系統之研發。本研究所建立之顯微成像系統，係基於全像術理論的光學原理之設計，不需透鏡元件，僅利用近同調之光源與影像感測器，並採用同軸架構以簡化設備需求，實現系統之微型化，可達到單次擷取近30 mm²之視場範圍並同時獲得影像強度與相位資訊，這相當於傳統顯微鏡20倍物鏡下的8倍範圍。在影像還原方面，本研究創新提出了一種通過建立影像遮罩以過濾出重要資訊之演算方式，成功解決了同軸全像術在影像重建過程中容易出現的偽影雜訊。此外，也進一步結合深度學習模型取代全像術的數學演算還原過程，提升影像與相位還原之速度與品質，並整合了雜訊消除與解析度提升之模型，進一步提高了還原影像的質量。在通過不同樣品的影像還原測試，最佳化此模型架構並進行可行性和有效性等驗證。最後採用單板電腦進行使用者介面之開發與整合，提升了本研究之實用性，並為未來的擴展應用提供了良好的基礎。本研究的成果展示了無透鏡全像顯微系統與人工智慧技術結合的巨大潛力，特別是在簡化光學設備、提高影像質量以及實現系統微型化方面的應用價值。未來，我們期待能夠進一步優化系統，加入更多先進的功能和應用，使其在各個領域中發揮更大的作用。 ;This study aims to explore the design of an optical system based on holography theory, utilizing a coaxial configuration to simplify equipment requirements and achieve miniaturization. Traditional holography requires extensive and complex optical equipment. In contrast, our designed system only needs a near-coaxial light source and sensor to capture an image area of nearly 30mm² in a single shot, equivalent to eight times the area under a 20x objective lens of a conventional microscope. In terms of image reconstruction, we innovatively proposed an algorithm that filters critical information through image masking based on the traditional holography reconstruction method (ASP), successfully solving the concentric circle issue in coaxial holography image reconstruction. Additionally, this study integrates the latest artificial intelligence technology by employing deep learning models to replace the mathematical theory of holography for image reconstruction. Through tests with different samples, the feasibility and effectiveness of this model have been demonstrated. We also combined noise reduction and resolution enhancement models, further improving the quality of reconstructed images. The comprehensive application of these technologies allows us to significantly enhance the clarity and accuracy of the images. The results of this study demonstrate the great potential of combining lensless holographic microscopy systems with artificial intelligence technology, especially its application value in simplifying optical equipment, improving image quality, and achieving system miniaturization. In the future, we look forward to further optimizing the system and adding more advanced functions and applications to make it play a more significant role in various fields. Keywords: holography, deep learning, image reconstruction, image quality enhancement, miniaturized system.
显示于类别:	[生物醫學工程研究所 ] 博碩士論文

文件中的档案:

档案	描述	大小	格式	浏览次数
index.html		0Kb	HTML	46	检视/开启

在NCUIR中所有的数据项都受到原著作权保护.

社群 sharing

数据加载中.....