工業4.0的時代，大量工件的自動生產已經成為趨勢；生產的同時，為了生產線的品質維持，需要一套可以檢測工件的自動化系統。如今，在檢測工件領域，多數選用自動化影像辨識，而深度學習在其中發揮很大的功用：透過光學系統與電腦中深度學習的程式進行連動，除了可以自動檢測外，也提升生產線效率。
然而，目前主流的自動檢測仍存在幾點限制：檢測用影像常須對整張影像處理，並且限定影像類屬單一類別，無法針對局部處理與偵測，所使用演算法也常需人為調整參數。
本研究針對上述問題，提出建立在自動光學檢測基礎上的多重區域性檢測技術，並且實踐自動調整演算法參數的能力，以及建立後續的區域瑕疵檢測，以實現更為有彈性的智慧化檢測。本研究使用技術包括機器學習中的分群法、深度學習以及深度強化學習，同時也使用影像處理的相關技術。
本研究對象為工業用的標籤紙張。透過影像處理與多項機器學習演算法的操作，可以實現影像的自動化區域分割、二維空間區域辨識資訊，以及區域影像內瑕疵驗證的能力。針對印刷的標籤進行光學檢測，驗證標籤上印刷的字元與特定的非字元圖章是否有印刷缺陷。本研究測試5種不同的模板，分別印製100張標籤影像，在標記瑕疵處後進行演算法驗證，最後結果在以標籤張數為單位的檢測率99 %、並且有1 %的過殺率。未來預期透過調整演算法，進一步發展區域辨識的研究，並且更有利於工業上發展。

In the era of Industry 4.0, the automated production of a large number of workpieces has become a trend. At the same time of production, in order to maintain the quality of the production line, an automated inspected system to test the workpieces is also needed. Nowadays, in the field of workpiece detection, usually automated image detection is chosen, and deep learning plays a big role in the application. By linking the optical system with the deep learning algorithm, not only automatic detection is available, but also the efficiency of the production line is improved.
However, there are still some shortcomings in the current main automatic inspection technique: First, the detected image is processed globally; second, the image category is usually limited to a single category. Also, it’s usually unavailable for local inspection and detection, while the related algorithm parameters are artificially adjusted.
In view of the above problems, this study proposes a multi-regional detection technology, which is based on automatic optical detection, and practice the ability to automatically adjust algorithm parameters, as well as establish subsequent regional defect detection, in order to achieve more flexible intelligent detection. This study uses techniques including unsupervised clustering in machine learning, deep learning and deep reinforcement learning. Also, digital image processing technique is used.
The current target of this study is the industrial label paper. Through the operation of multiple algorithms, the ability of automated image segmentation, two-dimensional space recognition, and preliminary verification of regional image defects are practiced. Perform optical inspection on the printed label to verify whether the printed characters and specific non-character marks on the label have printing defects. In this study, 5 different templates were tested, 100 images were printed respectively, and the algorithm was verified after marking the defects. The final detection rate is 99 % in the unit of label number, with 1 % over-killing rate. In the future, it is expected to adjust the algorithm, the research on regional identification will be further developed, and it will be more conducive to industrial development.

目錄
摘要 I
Abstract II
致謝詞 IV
目錄 VII
圖目錄 X
表目錄 XV
第一章 緒論 1
1.1 前言 1
1.2 研究動機 1
第二章 文獻回顧 3
2.1 自動光學檢測 3
2.1.1 概述 3
2.1.2 自動光學檢測技術硬體更新 7
2.1.3 自動光學檢測技術軟體更新 9
2.2 數位影像處理 12
2.2.1 概述 12
2.2.2 閥值化 13
2.2.3 邊緣特徵 15
2.2.4 噪聲 16
2.2.5 連通分量標記 17
2.2.6 影像輪廓學 19
2.2.7 影像特徵點 21
2.3 機器學習分群法 23
2.4 深度學習 28
2.4.1 多層神經網路 28
2.4.2 捲積神經網路 32
2.4.3 循環神經網路 36
2.4.4 光學字元辨識 39
2.5 深度強化學習 43
2.5.1 概述 43
2.5.2 值函數 46
2.5.3 策略梯度 48
2.5.4 演員-評論家 49
2.5.5 半無監督式分群法 52
第三章 研究方法 55
3.1 實驗設備硬體架構 55
3.2 檢測流程 60
3.3 影像預處理 63
3.4 影像分割 64
3.5 光學字元辨識與驗證 67
3.6 圖章辨識與驗證 71
3.7 半無監督式分群法模型 75
第四章 研究結果與討論 82
4.1 檢測結果單位 82
4.2 半無監督式分群法實驗結果 84
4.3 影像相減驗證光學字元瑕疵結果 90
4.4 光學字元驗證結果 92
4.5 圖章驗證結果 97
4.6 演算法耗費時間統計 98
4.7 半無監督式分群法調正參數比較 101
4.8 檢測錯誤子影像討論 103
第五章 結論 105
5.1 研究貢獻 106
5.2 未來展望 106
第六章 特別感謝 108
參考文獻 109

[1] Y. W. Lin, "Automatic defects classification from AOI images based on image processing and deep learning." National Tsing Hua University, Taiwan, 2018.
[2] T. S. Newman, and A. K. Jain, "A survey of automated visual inspection." Computer Vision and Image Understanding, vol. 61, issue 2, pp. 231-262, 1995.
[3] C. K. Cheng, “Formulation of Multi-Light Source Lighting Strategy Based on Reinforcement Learning to Detect Multi-Type Defect Area.” National Tsing Hua University, Taiwan, 2019.
[4] H. Rau, and C. H. Wu, "Automatic optical inspection for detecting defects on printed circuit board inner layers." International Journal of Advanced Manufacturing Technology, vol. 25, pp. 940-946, 2005.
[5] J. F. Borba and J. Facon, "A printed circuit board automated inspection system." Proceedings of the 38th Midwest Symposium on Circuits and Systems, pp. 69-72, 1995.
[6] J. Nakamura, Image sensors and signal processing for digital still cameras. published by CRC Press, Taylor & Francis Group, Boca Raton, 2006.
[7] A. Novini, "Fundamentals Of Machine Vision Lighting." Proc. SPIE 0728, Optics, Illumination, and Image Sensing for Machine Vision, pp. 84-92, 1987.
[8] Y. H. Won, H. Joo, and J. S. Kim, "Classification of Defects in the Polarizer of Display Panels using the Convolution Neural Network (CNN)." International Journal of Computing, Communications & Instrumentation Engineering, vol. 4, issue 1, pp. 139-142, 2017.
[9] P. Wei, C. Liu, M. Liu, Y. Gao, and H. Liu, "CNN-based reference comparison method for classifying bare PCB defects." The Journal of Engineering, issue 16, pp. 1528-1533, 2018.
[10] O. Russakovsky, J. Deng, H. Su, J. Krause, S. Satheesh, S. Ma, Z. Huang, A. Karpathy, A. Khosla, M. Bernstein, A. C. Berg, and L. L. Fei-Fei, “ImageNet Large Scale Visual Recognition Challenge.” International Journal of Computer Vision, pp. 1-37, 2014.
[11] K. Simonyan, and A. Zisserman, "Very deep convolutional networks for large-scale image recognition." International Conference on Learning Representations, 2015.
[12] K. He, X. Zhang, S. Ren, and J. Sun, "Deep residual learning
for image recognition." IEEE Conference on Computer Vision and Pattern Recognition, pp. 770-778, 2016.
[13] N. Otsu, "A threshold selection method from gray-level histograms." IEEE Transactions on Systems, Man, and Cybernetics, vol. 9, no. 1, pp. 62-66, 1979.
[14] H. C. Hsu, "3-Dimensional Defect Inspection of Workpieces Profile by Digital Image Processing." National Tsing Hua University, Taiwan, 2017.
[15] D. Y. Huang, T. W. Lin, and W. C. Hu, "Automatic multilevel thresholding based on two-stage Otsu's method with cluster determination by valley estimation." International Journal of Innovative Computing, Information & Control, vol. 7, no.10, pp. 5631-5644, 2011.
[16] V. Jumb, M. Sohani, and A. Shrivas, "Color Image Segmentation Using K-means Clustering and Otsu’s Adaptive Thresholding." International Journal of Innovative Technology and Exploring Engineering, vol. 3, issue 9, pp. 72-76, 2014.
[17] J. B. MacQueen, "Some Methods for classification and Analysis of Multivariate Observations." Proceedings of 5th Berkeley Symposium on Mathematical Statistics and Probability, University of California Press, pp. 281-297, 1967.
[18] Wikipedia-Sobel operator. (June 2019). Available: https://en.wikipedia.org/wiki/Sobel_operator
[19] J. Canny, "A computational approach to edge detection." IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 8, no. 6, pp. 679-698, 1986.
[20] A. K. Samantaray, and P. Mallick, "Decision Based Adaptive Neighborhood Median Filter." Procedia Computer Science, vol. 48, pp. 222-227, 2015.
[21] R. C. Gonzalez, and R. E. Woods, Digital Image Processing 3rd edition. published by Pearson Education Ltd., London, UK, 2007.
[22] E. Rublee, V. Rabaud, K. Konolige, and G. R. Bradski, "ORB: An efficient alternative to SIFT or SURF." International Conference on Computer Vision, vol. 11, pp. 2564-2571, 2011.
[23] P. V. C. Hough, "Method and means for recognizing complex patterns." U.S. Patent 3,069,654, 1962.
[24] R. O. Duda and P. E. Hart, "Use of the Hough Transformation to Detect Lines and Curves in Pictures." Communications of the ACM, vol. 15, pp. 11-15, 1972.
[25] S. Shukla, and Naganna S, "A Review on K-means data clustering approach." International Journal of Information & Computation Technology, vol. 4, no. 17, pp. 1847-1860, 2014.
[26] M. Ester, H.-P. Kriegel, J. Sander, and X. W. Xu, "A Density-Based Algorithm for Discovering Clusters in Large Spatial Databases with Noise." Proceedings of 2nd International Conference on KDD, pp. 226-231, 1996.
[27] F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, and E. Duchesnay, "Scikit-learn: Machine Learning in Python." Journal of Machine Learning Research, no. 12, pp. 2825-2830, 2011.
[28] S. R. Feng, and W. J. Xiao, "An improved DBSCAN clustering Algorithm." Journal of China University of Mining & Technology, vol. 37, no. 1, 2008.
[29] M. Copeland, “What’s the Difference Between Artificial Intelligence, Machine Learning and Deep Learning ?” https://blogs.nvidia.com/blog/2016/07/29/whats-difference-artificial-intelligence-machine-learning-deep-learning-ai/, July 29, 2016.
[30] W. S. McCulloch, and W. Pitts, "A logical calculus of the ideas immanent in nervous activity." The Bulletin of Mathematical Biophysics, vol. 5, issue 4, pp. 115-133, 1943.
[31] M. R. G. Meireles, P. E. M. Almeida, and M. G. Simoes, "A comprehensive review for industrial applicability of artificial neural networks." Transactions on Industrial Electronics, vol. 50, no. 3, pp. 585-601, 2003.
[32] D. E. Rumelhart, G. E. Hinton, and R. J. Williams, "Learning representations by back-propagating errors." Nature, vol. 323, pp. 533-536, 1986.
[33] C. Cortes, and V. Vapnik, "Support-vector networks." Machine Learning, vol. 20, issue 3, pp. 273-297, 1995.
[34] G. E. Hinton, S. Osindero, and Y. W. Teh, "A fast learning algorithm for deep belief nets." Neural Computation, vol. 18, no. 7, pp. 1527-1554, 2006.
[35] A. Krizhevsky, I. Sutskever, and G. Hinton, "Imagenet classification with deep convolutional neural networks." Neural Information Processing Systems, vol. 1, pp. 1097-1105, 2012.
[36] Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner, "Gradient-based learning applied to document recognition." Proceedings of the IEEE, vol. 86, pp. 2278-2324, 1998.
[37] W. Brendel and M. Bethge, "Approximating CNNs with Bag-of-local-Features models works surprisingly well on ImageNet." International Conference on Learning Representations, 2019.
[38] Y. Bengio, P. Simard, and P. Frasconi, "Learning long-term dependencies with gradient descent is difficult." IEEE Transactions on Neural Networks, vol. 5, no. 2, pp. 157-166, 1994.
[39] S. Hochreiter, and J. Schmidhuber. "Long short-term memory." Neural computation, vol. 9, issue 8, pp. 1735-1780, 1997.
[40] G. Nagy, T. A. Nartker, and S. V. Rice. "Optical Character Recognition: An illustrated guide to the frontier." Proc. SPIE 3967, Document Recognition and Retrieval VII,1999.
[41] Ramalingeswara Rao K V, Bhaskara Rao N, Ramesh Babu D R, Deeksha Goplani, and Anusha Pai. "Connected Component Analysis
for Character Feature Extraction." Proceedings of National Conference on Innovations in Electrical, Electronics and Computer Science Engineering, pp. 18-21, 2013.
[42] F. Zelic, and A. Sable, "A comprehensive guide to OCR with Tesseract, OpenCV and Python." https://nanonets.com/blog/ocr-with-tesseract/ (Available in June 2020), 2019.
[43] R. Smith, "An Overview of the Tesseract OCR Engine." Proc. International Conference on Document Analysis and Recognition, 2007.
[44] X. Guo and Y. C. Fang, 最新人工智慧應用：用強化學習快速上手AI. 佳魁資訊股份有限公司, 2018.
[45] C. J. Watkins and P. Dayan, "Q-learning." Machine learning, vol. 8, no. 3-4, pp. 279-292, 1992.
[46] V. Mnih, K. Kavukcuoglu, D. Silver, A. A. Rusu, J. Veness, M. G. Bellemare, A. Graves, M. Riedmiller, A. K. Fidjeland, G. Ostrovski, S. Petersen, C. Beattie, A. Sadik, I. Antonoglou, H. King, D. Kumaran, D. Wierstra, S. Legg, and D. Hassabis, "Human-level control through deep reinforcement learning." Nature, vol. 518, no. 7540, pp. 529-541, 2015.
[47] H. Van Hasselt, A. Guez, and D. Silver, "Deep Reinforcement Learning with Double Q-Learning." Association for the Advancement of Artificial Intelligence, vol. 2, pp. 2094-2100, 2016.
[48] J. Schulman, S. Levine, P. Abbeel, M. Jordan, and P. Moritz, "Trust region policy optimization." International Conference on Machine Learning, pp. 1889-1897, 2015.
[49] T. P. Lillicrap, J. J. Hunt, A. Pritzel, N. Heess, T. Erez, Y. Tassa, D. Silver, and D. Wierstra, "Continuous control with deep reinforcement learning." International Conference on Learning Representations, 2016.
[50] I. Grondman, L. Busoniu, G. Lopes, and R. Babuska. "A survey of actor-critic reinforcement learning: standard and natural policy gradients." IEEE Transactions on Systems, Man, and Cybernetics, Part C: Applications and Reviews, Institute of Electrical and Electronics Engineers, vol. 42, issue 6, pp.1291-1307, 2012.
[51] G. E. Uhlenbeck, and L. S. Ornstein, "On the theory of Brownian Motion." Physical Review, vol. 36, issue 5, pp. 823–841, 1930.
[52] S. Mannor, I. Menache, A. Hoze, and U. Klein, "Dynamic Abstraction in Reinforcement Learning via Clustering." Proceedings of the 21st International Conference of Machine Learning, pp. 560-567, 2004.
[53] S. Bose, and M. Huber, "Semi-Unsupervised Clustering Using Reinforcement Learning." Proceedings of the Twenty-Ninth International Florida Artificial Intelligence Research Society Conference, pp. 150-153, 2016.
[54] X. Guo, and Y. Zhai, "K-Means Clustering based Reinforcement Learning Algorithm for Automatic Control in Robots." International Journal of Simulation, System, Science and Technology, vol. 17, pp. 1-6, 2016.