臺灣博碩士論文加值系統

今天，我們對工業4.0的概念以及高品質的產品並不太陌生。 然而，避免生產過程中出現的缺陷是一項挑戰。 因此，檢測和分類這些缺陷是一項重大挑戰。
光學透鏡用於各種目的。 獲得合格的工業、醫療、教育和交通光學鏡片非常重要，例如智能手機鏡片、隱形眼鏡、車燈鏡片、激光束校正鏡片等。然而，即使有微小的缺陷，它也會影響 產品質量問題，使得生產過程中出現錯誤。 因此，設計一套自動光學鏡片檢測系統是必不可少的要求。
日常生活智能手機鏡片、隱形眼鏡、車燈鏡片或一些用於校正激光束的光學鏡片。儘管只有很小的缺陷，也會影響到產品的品質，使用過程中出現錯誤。因此，本研究提出了一種使用自適應奇異值分解離散餘弦變換 (ASVDD) 和對比度受限自適應直方圖均衡 (CLAHE) 方法對鏡片上存在的缺陷進行識別和分類的過程中提高鏡片圖像質量的方法。照明實驗室（CIELAB）顏色通道和其他一些方法。該方法對應於圖像的不同部分，並使用它們重新分配圖像密度值，在 CLAHE 參數的情況下，“分佈”用於定義表達式的形狀。比比較然後均衡自適應直方圖均衡 (AHE) 產生更好質量結果的圖形。最後使用一些深度學習方法來評估和呈現結果。缺陷檢測率超過98.69%，結果比原圖提高5.11%。因此，結果表明，將該方法應用於圖像增強過程可以提高圖像質量，從而對產品的質量做出準確的決策。
關鍵詞 - 深度學習、離散餘弦變換、圖像增強、自動光學檢測、奇異值分解。

Today, we are not too unfamiliar with the concept of Industry 4.0 along with high-quality products. However, it is challenging to avoid defects appearing in the production process. Therefore, detecting and classifying these defects are significant challenges.
The optical lenses are used for various purposes. It is so important to obtain qualified optical lenses for industry, medicine, education, and transportation, such as smartphone lenses, contact lenses, lenses for car lights, lenses for correcting laser beams, etc. However, even with tiny defects, it also affects the quality of the product, making the production process have errors. Therefore, designing an automatic optical lens inspection system is an essential requirement.
This dissertation proposes an automatic optical lens inspection system based on computer vision and deep learning. First, the optical lens sample image is captured by an Olympus BX40 microscope. Next, the captured image is enhanced by using adaptive singular value decomposition in discrete cosine domain (ASVDD) and contrast limited adaptive histogram equalization (CLAHE) methods. Finally, the enhanced image is input to the YOLOv5 deep learning model for defect detection. The experiments show that the defect detection rate is over 98.69%, higher than 5.11% compared to the original image. To demonstrate the effectiveness of our proposed method, we also compare it with state-of-art techniques such as ASVDF (adaptive singular value decomposition in the Fourier domain) and ASVDW (adaptive singular value decomposition in the wavelet domain). The results show that our proposed method has more advantages in practical applications, both in terms of error detection efficiency and faster computation time.
Index Terms - Deep learning, discrete cosine transform, image enhancement, automated optical inspection, singular value decomposition.

LIST OF FIGURES XII
LIST OF TABLES XIII
ABBREVIATIONS XIV
CHAPTER 1 ORGANIZATION 1
1.1 Motivation 1
1.2 Dissertation Organization 2
CHAPTER 2 INTRODUCTION 3
2.1 Overview of the Optical Lens Samples 3
2.2 The Image Acquisition System 5
2.3 Object Recognition System Overview 7
2.4 Optical Lens Defect Image Database 9
2.5 Summary 10
CHAPTER 3 IMAGE ENHANCEMENT FOR OPTICAL LENS INSPECTION SYSTEM 11
3.1 Adaptive Singular Value Decomposition in DCT Domain 11
3.2 Contrast Limited Adaptive Histogram Equalization 16
3.3 Non-Local Means Denoising 23
3.6 The Proposed Framework Image Enhancement for Optical Lens Inspection System 25
3.4 Summary 26
CHAPTER 4 EXPERIMENTAL RESULT AND DISCUSSION 28
4.1 Defect Image Enhancement Experiment 28
4.2 Defect Recognition based on Deep Learning Model Discussion 32
4.3 Summary 36
CHAPTER 5 CONCLUSION AND FUTURE WORK 37
5.1 Conclusion 37
5.2 Future Work 38
REFERENCES 40
BIOGRAPHY 42

[1]Y-Q. Lin, Y-C. Xiang, Y-X. Lin, and J. Yu, “Defect detection system for optical element surface based on machine vision,” 2019 2nd International Conference on Information Systems and Computer Aided Education (ICISCAE), Dalian, China, Sept. 2019.
[2]H. C. Andrews and C. L. Patterson, “Singular value decompositions and digital image processing,” IEEE Trans. Acoust. Speech. Signal Process, vol. 24, no. 1, pp. 26-53, 1976.
[3]H. C. Andrews and C. L. Patterson, “Singular value decomposition and digital image processing,” IEEE Trans. Acoustic, Speech, and Signal Proc, vol. 24, 1976.
[4]J. Liu, S. Chen, and X. Tan, “Fractional order singular value decomposition representation for face recognition,” Pattern Recognit, vol. 41, pp. 378-395, 2008.
[5]Y. Tian, T. Tan, Y. Wang, and Y. Fang, “Do singular values contain adequate information for face recognition,” Pattern Recognit, vol. 36, pp. 649–655, 2003.
[6]J.-W. Wang, J.-S. Lee, and W.-Y. Chen, “Face recognition based on projected color space with lighting compensation,” IEEE Signal Process. Lett, vol. 18, pp. 567-570, 2011.
[7]J. Zhou and P. Chen, “Generalized Discrete Cosine Transform,” 2009 Pacific-Asia Conference on Circuits, Communications and Systems, pp. 449-452, Sept. 2009.
[8]C-C. Tseng, and S-L. Lee, “A Low-Light Color Image Enhancement Method on CIELAB Space,” 2018 IEEE 7th Global Conference on Consumer Electronics (GCCE), Nara, Japan, Dec. 2018.
[9]V. Stimper, S. Bauer, R. Ernstorfer, B. Schölkopf, and R. P. Xian, “Multidimensional Contrast Limited Adaptive Histogram Equalization,” IEEE Access, vol. 7, pp. 165437 – 165447, Nov. 2019.
[10]S. H. Majeed and N. A. M. Isa, “Adaptive Entropy Index Histogram Equalization for Poor Contrast Images,” IEEE Access, vol. 9, pp. 6402 - 6437, Jul 1997.
[11]G. Yadav, S. Maheshwari, and A. Agarwal, “Contrast limited adaptive histogram equalization based enhancement for real time video system,” 2014 International Conference on Advances in Computing, Communications and Informatics (ICACCI), Delhi, India, Dec. 2014.
[12]D. V. D. Ville and M. Kocher, “SURE-Based Non-Local Means,” IEEE Signal Processing Letters, vol. 16, no. 11, Nov 2009.
[13]J.- W. Wang, N. T. Le, J.- S. Lee, and C.- C. Wang, “Color face image enhancement using adaptive singular value decomposition in Fourier domain for face recognition,” Pattern Recognit. vol. 57, pp. 31-49, Sept. 2016.
[14]J.- W. Wang, N. T. Le, J.- S. Lee, and C.- C. Wang, “Illumination compensation for face recognition using adaptive singular value decomposition in the wavelet domain,” Inf. Sci. vol. 435, pp. 69-93, Apr. 2018.