臺灣博碩士論文加值系統

透鏡使用範圍寬泛，從日常的手機鏡頭、研究用的顯微鏡以及天文望遠鏡，各種需要變焦與成像的東西都需要用到透鏡。因此，透鏡對於一些會影響成像的瑕疵可以說不能容忍，對此需要強力且有效的瑕疵檢測系統。有鑑於此，本研究創新設計透鏡專用光學檢測架構以及客製化光場，以應用於瑕疵精密檢測。凹凸透鏡共30片用於實驗並進行測試，初檢採用本實驗室團隊創新設計的自適應奇異值分解(ASVD)技術來強化影像品質凸顯瑕疵所在，以篩檢出滴點、灰塵及滑痕；後續複檢則採用傅利葉域自適應奇異值分解(ASVDF)複檢積液跟滴痕。實驗結果證明，本論文提出的研究架構可以有效提升透鏡瑕疵的捕獲數。在未經強化的100張500 × 500像素的原始圖像上可以平均抓取33個瑕疵。ASVD強化後的圖像可以平均抓取48個瑕疵，ASVDF強化後的圖像可以平均抓取56個瑕疵。ASVD平均單張強化時間為0.28秒，而ASVDF平均單張強化時間為3.21秒。實驗結果證明，本論文提出的創新系統，可以有效的提升透鏡瑕疵捕獲數。

Lenses have a wide range of applications, from everyday camera lenses on smartphones to microscopes used in research and astronomical telescopes. Various objects that require zooming and imaging rely on lenses. Therefore, lenses do not tolerate defects that could affect imaging quality. A robust and efficient defect detection system is necessary. In light of this, this study innovatively designs a specialized optical inspection framework for lenses and customizes the light field for precise defect detection.
A total of 30 convex-concave lenses are used in the experiments and tests. Initial inspection uses the adaptive singular value decomposition (ASVD) technique innovatively designed by our laboratory team to enhance the images and highlight defect locations. This is used to screen for droplets, dust, and smudges. Subsequent re-inspection employs the adaptive singular value decomposition in the Fourier domain (ASVDF) for further inspection of liquid residue and droplet traces.
Experimental results demonstrate that the proposed research framework effectively enhances the detection of lens defects. In the unenhanced original images of 100 samples with 500 × 500 pixels, an average of 33 defects per picture is detected. Images enhanced with ASVD capture an average of 48 defects per picture, while images enhanced with ASVDF capture an average of 56 defects per picture. The average single-frame enhancement time for ASVD is 0.28 seconds, while for ASVDF, it is 3.21 seconds. The experimental results validate the efficacy of the innovative system proposed in this study, effectively increasing the number of lens defects captured.

中文摘要 I
Abstract II
致謝 III
目錄 IV
圖目錄 VI
表目錄 VIII
第一章 緒論 1
1.1簡介 1
1.2研究背景與動機 1
1.3相關文獻回顧 2
1.4透鏡檢測系統的問題描述及圖像處理檢測系統架構 3
1.5章節概要 5
第二章 透鏡取像系統與瑕疵 6
2.1簡介 6
2.2相機鏡頭與影像拍攝設定 6
2.3光源 7
2.4透鏡支架 9
2.5拍攝架構 11
2.6 瑕疵介紹 12
2.7 拍攝要點 16
第三章 程式流程與圖像處理技術 18
3.1簡介 18
3.2程式流程 18
3.3影像前置處裡 20
3.4自適應奇異值分解（ASVD） 22
3.5傅立葉域自適應奇異值分解（ASVDF） 24
3.6 能量分析（Energy Analysis） 26
第四章 實驗與結果 28
4.1拍攝系統之實驗 28
4.1.1光源選擇實驗 28
4.1.2取像架構 31
4.2圖像處理系統實驗 33
4.2.1系統規格 33
4.2.2圖像強化技術驗證實驗 34
4.2.3 ASVD與ASVDF強化技術處理時間比較 38
4.2.4 ASVD與ASVDF強化技術圈選能力比較 39
第五章 結論與未來工作 43
文獻參考 44

[1]Eugene Hecht, 2016, Optics, 5/e, Pearson FT Press.
[2]Joachim Wittig, 2013, Ernst Abbe, Vieweg+Teubner Verlag Wiesbaden.
[3]鄭煒穎，深度學習應用於護目鏡瑕疵檢測與分類，國立高雄科技大學，光電工程研究所，碩士論文，2021。
[4]Richard O. Duda, and Peter E. Hart, “Use of the Hough transformation to detect lines and curves in pictures,” Communications of the ACM, Vol. 15, issue 1, pp. 11-15, Jan. 1972.
[5]H. K. Yuen, John P. Princen, J. Illingworth, and J. Kittler, “Comparative study of Hough Transform methods for circle finding,” ScienceDirect Image and Vision Computing, Vol. 8, Issue 1, pp. 71-77, Feb. 1990.
[6]Jing-Wein Wang, Jiann-Shu Lee, and Wen-Yuan Chen, “Face recognition based on projected color space with lighting compensation,” IEEE Signal Processing Letters, Vol.18, Oct. 2011.
[7]Jing-Wein Wang, Ngoc Tuyen Le, Jiann-Shu Lee, and Chou-Chen Wang, “Color face image enhancement using adaptive singular value decomposition in Fourier domain for face recognition,” Pattern Recognition, Vol. 57, pp. 31-49, Sep. 2016.
[8]吳家華，應用自適應能量分析於自動光學檢測，國立高雄科技大學，光電與通訊工程研究所，碩士論文，2017。
[9]Basler 官方網站acA4112-30uc- Basler ace規格, https://www.baslerweb.com/en/products/cameras/area-scancameras/ace/aca4112-30uc/。
[10]韓上原，影像增強應用於深度學習透鏡光電檢測系統，國立高雄科技大學，電子工程系，碩士論文，2022。
[11]Jing-Wein Wang, Ngoc Tuyen Le, Chou-Chen Wang, and Jiann-Shu Lee, “Enhanced ridge structure for improving fingerprint image quality based on a wavelet domain,” IEEE Signal Processing Letters, Vol. 22, No. 4, Apr. 2015.
[12]Zhou Wang, Alan C. Bovik, Hamid R. Sheikh, and Eero P. Simoncelli, “Image quality assessment: from error visibility to structural similarity,” IEEE Transactions on Image Processing, Vol. 13, No. 4, Apr. 2004.
[13]深入淺出Yolo系列之Yolov5核心基礎知識完整講解，https://zhuanlan.zhihu.com/p/172121380。
[14]Responding to the Controversy about YOLOv5，https://blog.roboflow.com/yolov4-versus-yolov5/.