本論文利用多角度攝影鏡頭藉由影像處理與深度學習之相關技術結合，為設計一套基於剎車來令片表面檢測系統因生產剎車來令片過程中製程影響導致來令片周圍有缺角與表面有膨脹之問題。本論文研究過程中主要分為三個部分，第一部分為蒐集檢測物來令片建置資料集，透過影像處理將檢測物之特徵擷取與標註分類，經由程式控制攝像頭取代人力拍攝；第二部分為深度學習將前者蒐集完成的資料集輸入神經網路中進行訓練，透過調整優化函數以及損失函數來達到最佳化模型，此技術可以利用深度學習，讓電腦具備與人一樣的分類技術；最後一個部分為多角度攝影鏡頭之邊緣裝置佈署，將兩個二維投影面之座標同步來達到多角度瑕疵檢測。本論文主要之目的為有效運用影像處理解決瑕疵與正樣本數不足等問題與多角度檢測結合深度學習模型來判別剎車來令片表面之瑕疵。

In this thesis, image processing (OpenCV) and deep learning (Tensorflow) based on dual-camera techniques are used to detect the surface defects of brake pads in real-time. Due to manufacturing process of the brake pads, there problems of missing corners around the pads and swelling of the surface. This thesis is mainly divided into three parts. Firstly, it is to collect images and create dataset for image processing to extract the features of the detected objects and to classify the images for the next stage. classified for neural network for training. The optimized model is achieved by adjusting the optimization function and the loss function. Finally, the devices are set up for the multi-camera coordinate synchronization of 2D coordinate synchronization of 2D projection surface to achieve multi-angle defect detection. This thesis is to use image processing to detect defects effectively; to solve the problems of insufficient number of positive samples; to use multi-angle detection with deep learning model to identify the defects on the surface of brake pads.

摘要......i
Abstract......ii
誌謝......iii
目錄......iv
表目錄......vi
圖目錄......vii
第一章 緒論......1
1.1研究動機與目的......1
1.2 文獻回顧......3
1.3 論文架構......5
第二章 論文相關技術分析與文獻探討......6
2.1 影像處理......6
2.1.1 影像前處理......6
(1) 灰階處理......6
(2) 幾何轉換......7
(3) 圖像二值化......8
(4) 圖像平滑......9
(5) 形態學變換......10
(6) 邊緣偵測......11
(7) 輪廓偵測......13
2.2 人工神經網路......14
2.2.1 人工神經網路架構......15
(1) 單層感知器(Single Layer Perceptron)......15
(2) 多層感知器(Multilayer Perceptron, MLP)......15
(3) 激活函數(Activation Function)......16
2.2.2 反向傳播算法......18
2.3 卷積神經網路......20
(1) 卷積層......20
(2) 池化層......21
(3) 扁平層......22
(4) 全連接層......22
2.4 物件偵測模型架構......23
(1) 單階段式......23
(2) 二階段式......26
第三章 研究架構與實驗方法......29
3.1 研究架構介紹......29
3.2 研究資料庫樣本蒐集......30
3.3 影像資料前處理......32
3.4 建構深度學習模型......34
(1) 深度學習中運行CPU與GPU......35
(2) 遷移式學習......36
(3) EfficientNet模型架構......38
(4) 物件偵測模型設計......42
3.5 佈署多角度鏡頭與模型至邊緣裝置......44
第四章 實驗結果與討論......47
4.1影像二值化、模糊化與邊緣檢測之關係實驗結果......47
4.2邊緣區域與找尋最大輪廓之實驗結果......50
4.3資料庫樣本蒐集之實驗結果......51
4.4搭建模型之實驗結果......52
4.4模型最佳化實驗結果......54
4.5多角度相機基於二維定位實驗結果......58
4.6實瑕疵檢測之邊緣裝置佈署實驗結果......60
4.7實驗結果討論......61
第五章 結論與未來研究方向......63
5.1 結論......63
5.2 未來研究方向......64
參考文獻......65
Extended Abstract......67

[1]B. R. Mete and T. Ensari, "Flower Classification with Deep CNN and Machine Learning Algorithms,"2019 3rd International Symposium on Multidisciplinary Studies and Innovative Technologies (ISMSIT), Ankara, Turkey, 2019, pp. 1-5.
[2]Y. LeCun et al., "Backpropagation Applied to Handwritten Zip Code Recognition," Neural Computation, vol. 1, no. 4, pp. 541-551, Dec. 1989.
[3]Y. LeCun, K. Kavukcuoglu and C. Farabet, "Convolutional networks and applications in vision,"Proceedings of 2010 IEEE International Symposium on Circuits and Systems, Paris, 2010, pp. 253-256.
[4]Li Y, Huang H, Xie Q, Yao L and Chen Q, "Research on a Surface Defect Detection Algorithm Based on MobileNet-SSD," Applied Sciences,2018, pp. 1-17.
[5]Shu Y and Li B," Surface Defect Detection and Recognition Method for Multi-Scale Commutator Based on Deep Transfer Learning, " Arab J Sci Eng, 2021 Jul 30:1-12, pp.3-10.
[6]A. S. Olagoke, H. Ibrahim and S. S. Teoh, "Literature Survey on Multi-Camera System and Its Application," in IEEE Access, vol. 8, pp. 172892-172922, 2020, doi: 10.1109/ACCESS.2020.3024568.
[7]B. Rinner and W. Wolf, "An introduction to distributed smart cameras", Proc. IEEE, vol. 96, no. 10, pp. 1565-1575, Oct. 2008.
[8] R. Girshick, J. Donahue, T. Darrell and J. Malik, "Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation," 2014 IEEE Conference on Computer Vision and Pattern Recognition, 2014, pp. 580-587.
[9]R. Girshick, "Fast R-CNN," 2015 IEEE International Conference on Computer Vision (ICCV), 2015, pp. 1440-1448.
[10]S. Ren, K. He, R. Girshick and J. Sun, "Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks," in IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 39, no. 6, June 2017, pp. 1137-1149.
[11]K. He, G. Gkioxari, P. Dollár and R. Girshick, "Mask R-CNN," 2017 IEEE International Conference on Computer Vision (ICCV), 2017, pp. 2980-2988.
[12]J. Redmon, S. Divvala, R. Girshick and A. Farhadi, "You Only Look Once: Unified, Real-Time Object Detection," 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016, pp. 779-788.
[13]Liu, W. et al, " SSD: Single Shot MultiBox Detector, " In: Leibe, B., Matas, J., Sebe, N., Welling, M. (eds) Computer Vision – ECCV 2016. ECCV 2016. Lecture Notes in Computer Science ,vol 9905. Springer, 2016, pp. 21-37.
[14]李立宗，「OpenCV輕鬆入門:=面向Python」，電子工業出版社，2019年5月
[15]E. S. Gedraite and M. Hadad, "Investigation on the effect of a Gaussian Blur in image filtering and segmentation," Proceedings ELMAR-2011, 2011, pp. 393-396.
[16]W. Rong, Z. Li, W. Zhang and L. Sun, "An improved Canny edge detection algorithm," 2014 IEEE International Conference on Mechatronics and Automation, 2014, pp.577-582.
[17]Xin-Yi Gong, Hu Su, De Xu, Zheng-Tao Zhang, Fei Shen and Hua-Bin Yang. " An Overview of Contour Detection Approaches," International Journal of Automation and Computing, vol. 15, no. 6, 2018, pp. 656-672.
[18]B. Widrow and M. A. Lehr, "30 years of adaptive neural networks: Perceptron, madaline, and backpropagation, " Proceedings of the IEEE, 1990 ,Vol. 78, No. 9, pp. 1415–1442
[19]Madrid-Sánchez, J., Lázaro-Gredilla, M., Figueiras-Vidal and A.R., "A Single Layer Perceptron Approach to Selective Multi-task Learning, ", IWINAC 2007. Lecture Notes in Computer Science, 2007, pp 272–281.
[20]G. Singh and M. Sachan, "Multi-layer perceptron (MLP) neural network technique for offline handwritten Gurmukhi character recognition," 2014 IEEE International Conference on Computational Intelligence and Computing Research, 2014, pp. 1-5.
[21]Rumelhart, D., Hinton, G. & Williams. R, "Learning representations by back-propagating errors," Nature 323, 1986, 533–536.
[22]S. Dasiopoulou, V. Mezaris, I. Kompatsiaris, V. -. Papastathis and M. G. Strintzis, "Knowledge-assisted semantic video object detection," in IEEE Transactions on Circuits and Systems for Video Technology, vol. 15, no. 10, Oct. 2005, pp. 1210-1224.
[23]J. R. Uijlings, K. E. Sande, T. Gevers, and A. W. Smeulders , " Selective Search for Object Recognition. ", Int. J. Comput. Vision 104, (September 2013), 154–171.
[24]Gary Bradski and Adrian Kaehler, Learning OpenCV, Sebastopol, USA, 2017 pp.3-300.
[25]Jason Yosinski, Jeff Clune, Yoshua Bengio and Hod Lipson, "How transferable are features in deep neural networks? ", In Advances in Neural Information Processing Systems 27 (NIPS ’14),, Dec. 2014, pp. 3320-3328.
[26]Tan, M., & Le, Q, " Efficientnetv2: Smaller models and faster training," In International Conference on Machine Learning (PMLR) ,2021, pp. 10096-10106).
[27]J. Lee, J. Bang and S. -I. Yang, "Object detection with sliding window in images including multiple similar objects," 2017 International Conference on Information and Communication Technology Convergence (ICTC), 2017, pp. 803-806
[28]HOSANG, Jan; BENENSON, Rodrigo; SCHIELE, Bernt," Learning non-maximum suppression. " In: Proceedings of the IEEE conference on computer vision and pattern recognition. 2017. p. 4507-4515.