臺灣博碩士論文加值系統

本篇論文利用市售的不銹鋼板模擬經過切割程序後，其不銹鋼板表面可能會產生刮傷、鑿痕、汙點等不同的瑕疵，由不銹鋼板表面具有亮反光的特性，傳統上因採用的人工比對檢測法不但耗時費力，且容易因為高反光的干擾而出現檢測結果不同將使不良品流出，本研究使用現今進展相當成熟的深度學習，尤其是圖片識別與影像識別更是廣泛應用，通常圖片辨識與影像辨識都是從一般的光線加以識別，如何在高反光不銹鋼板上進行圖像瑕疵辨別，因此光線在圖像上面也是重要的議題。
而本篇採用 YOLOv3、SSD 300 架構為基底，修改其內部的卷積神經網路，透過不同參數進行不銹鋼板瑕疵偵測的網路架構，且有效提高偵測不銹鋼板的準確度。並透過圖像前處理標記樣本瑕疵相對位置，提高訓練前處理資料的動作，使其快速有效的處理大量的圖片樣本加以進行訓練。

This paper proposes simulate the cutting process after which uses commercially available of stainless steel plate. Its surface stainless steel plate may produce different flaws such as scratch, chisel, stain defects. The surface of stainless steel plate has bright reflective properties. The traditional manual comparison detection method which is not only time consuming and laborious, and is more for different detection results to appear due to the interference of high reflection, which will cause defective products to flow out.This study which uses progress of deep learning is fairly mature nowadays, especially picture recognition and image recognition. Usually picture recognition and image recognition are identified from ordinary light. How to identify image defects on highly reflective stainless steel plates, so light is also an important issue on the image.
Based on the YOLOv3、SSD 300 architecture, this thesis modifies its internal CNN a network architecture for flaw detection of stainless steel plate through different parameters,and effectively improve the accuracy of detecting stainless steel plate, and mark the relative position of sample defects through image preprocessing, and the action of processing the data before training can be improved, so that a large number of pictures can be quickly generated for training.

摘 要 i
Abstract ii
誌 謝 iv
目 錄 v
圖 目 錄 viii
表 目 錄 x
第1章 緒論 1
1.1. 研究動機與背景 1
1.2. 研究目的 2
1.3. 研究範圍與限制 2
1.4. 研究架構 3
第2章 文獻回顧 4
2.1. 機器學習 4
2.1.1. 學習分類模式 5
2.2. 類神經網路 6
2.3. 卷積神經網路 7
2.3.1. 卷積層 8
2.3.2. 池化層 9
2.3.3. 激勵函數 10
2.4. 深度學習 11
2.5. 深度學習框架 13
2.5.1. Keras 14
2.5.2. Tensorflow 14
2.6. 物件偵測 15
2.6.1. YOLOv1 15
2.6.2. YOLOv2 20
2.6.3. YOLOv3 22
2.6.4. SSD 24
第3章 研究方法 27
3.1. 不銹鋼影像與缺陷定義 27
3.1.1. 缺陷定義 28
3.2. 影像擷取 30
3.2.1. 實驗檢測環境 31
3.3. 訓練集樣本 33
3.3.1. 標記樣本 36
3.4. YOLOv3的訓練流程 43
3.4.1. YOLOv3模型訓練 44
3.5. SSD 300的訓練流程 45
3.5.1. SSD 300模型訓練 46
第4章 實驗結果與探討 47
4.1. YOLOv3模型訓練結果 47
4.1.1. YOLOv3模型分類正確率之驗證 49
4.2. SSD 300模型訓練結果 54
4.2.1. SSD 300模型分類正確率之驗證 56
4.3. 不銹鋼板瑕疵檢測比較結果 61
第5章 結論與未來展望 62
5.1. 結論 62
5.2. 未來展望 63
參考文獻 64

[1]劉曉薇、陳思翰，(2017).金屬二次加工自動化,光學檢測技術,智慧機械,機械資訊月刊.
[2]Rosenblatt,F.(1958).The perceptron: a probabilistic model for information
storage and organization in the brain.Psychological review, 65(6), 386.
[3]Gardner,M.W.,& Dorling,S.R.(1998).Artificial neural networks (the multilayer perceptron)—a review of applications in the atmospheric sciences. Atmospheric environment, 32(14-15), 2627-2636.
[4]Rumelhart,D.E.,Hinton,G.E.,& Williams,R.J.(1986).Learning representations by back-propagating errors. nature,323(6088),533.
[5]Hecht-Nielsen,R.(1992).Theory of the backpropagation neural network. In Neural networks for perception (pp.65-93).
[6]Quinlan,J.R.(1986).Induction of decision trees.Machine learning, 1(1),81-106.
[7]Cortes,C.,& Vapnik,V.(1995).Support-vector networks.Machine learning,20(3),273-297.
[8]LeCun,Y.,Bottou,L.,Bengio,Y.,& Haffner,P.(1998).Gradient-based learning applied to document recognition.Proceedings of the IEEE, 86(11),2278-2324.
[9]Hinton,G.E.,Osindero,S.,& Teh,Y.W.(2006).A fast learning algorithm for deep belief nets. Neural computation,18(7),1527-1554.
[10]Krizhevsky,A.,Sutskever,I.,& Hinton,G.E.(2012).ImageNet classification with deep convolutional neural networks.In Advances in neural information processingsystems (pp.1097-1105).
[11]Socher,R.,Bauer,J.,& Manning,C.D.(2013).Parsing with compositional vector grammars.In Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) (Vol. 1,pp.455-465).
[12]A.M. Legendre. Nouvelles méthodes pour la détermination des orbites des comètes, Firmin Didot, Paris, 1805. 「Sur la Méthode des moindres quarrés」 appears as an appendix.
[13]Mitchell, T.M. (1980). The need for biases in learning generalizations. CBM-TR 5-110, Rutgers University, New Brunswick, NJ.
[14]Jardins, M., and Gordon, D.F. (1995). Evaluation and selection of biases in machine learning. Machine Learning Journal, 5:1--17, 1995.
[15]Altman, N. S. An introduction to kernel and nearest-neighbor nonparametric regression. The American Statistician. 1992, 46 (3): 175–185.
[16]Salamin, Eugene, Computation of pi, Charles Stark Draper Laboratory ISS memo 74–19, 30 January, 1974, Cambridge, Massachusetts.
[17]Caruana, R. and Niculescu-Mizil, A.: An empirical comparison of supervised learning algorithms. Proceedings of the 23rd international conference on Machine learning, 2006.
[18]Broomhead, David H.; Lowe, David. Multivariable Functional Interpolation and Adaptive Networks (PDF). Complex Systems. 1988, 2: 321–355.
[19]Goodfellow, Ian J.; Pouget-Abadie, Jean; Mirza, Mehdi; Xu, Bing; Warde-Farley, David; Ozair, Sherjil; Courville, Aaron; Bengio, Yoshua.Generative Adversarial Networks. 2014.
[20]Haykin,Simon.9.Self-organizing maps.Neural networks - A comprehensive foundation 2nd. Prentice-Hall.1999.
[21]Carpenter,G.A. and Grossberg, S.,ART2: self-organisation of stable category recognition codes for analog input patterns, Applied Optics, 26(23), 4919-4930, 1987.
[22]LeCun,Y.,Bottou,L.,Bengio,Y.,& Haffner,P.(1998).Gradient-based learning applied to document recognition. Proceedings of the IEEE, 86(11), 2278-2324.
[23]Zeiler, M. D., & Fergus, R. (2014, September). Visualizing and understanding
convolutional networks. In European conference on computer vision (pp.818-833). Springer, Cham.
[24]Simonyan, K., & Zisserman, A. (2014). Very deep convolutional networks for largescale image recognition. arXiv preprint arXiv:1409.1556.
[25]Szegedy, C., Liu, W., Jia, Y., Sermanet, P., Reed, S., Anguelov, D., ... & Rabinovich, A. (2015, June). Going deeper with convolutions. Cvpr.
[26]He, K., Zhang, X., Ren, S., & Sun, J. (2016). Deep residual learning for image
recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 770-778).
[27]Lin, C. H., Liu, J. C., & Lee, K. Y. (2018). On Neural Networks for Biometric
Authentication Based on Keystroke Dynamics. Sensors and Materials, 30(3), 385396.
[28]S.J. Nowlan and G. E. Hinton. Simplifying neural networks by soft weight-sharing. Neural Computation, 4(4), 1992.
[29]V. Nair and G. E. Hinton. “Rectified linear units improve restricted boltzmann machines” . In Proc. 27th International Conference on Machine Learning, 2010.
[30]Y. Lecun, L. Bottou, Y. Bengio, and P. Haffner, “Gradient-based learning applied to document recognition,” Proceedings of the IEEE, vol. 86, no. 11, pp. 2278–2324, Nov. 1998.
[31]J. Deng, W. Dong, R. Socher, L. J. Li, K. Li, and L. Fei-Fei, “ImageNet: A large-scale hierarchical image database,” in 2009 IEEE Conference on Computer Vision and Pattern Recognition, 2009, pp. 248–255.
[32]M. D. Zeiler and R. Fergus, “Visualizing and Understanding Convolutional Networks,” arXiv:1311.2901 [cs], Nov. 2013.
[33]R. Girshick, J. Donahue, T. Darrell, and J. Malik, “Region-Based Convolutional Networks for Accurate Object Detection and Segmentation,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 38, no. 1, pp. 142–158, Jan. 2016.
[34]J. R. R. Uijlings, K. E. A. van de Sande, T. Gevers, and A. W. M. Smeulders, “Selective Search for Object Recognition,” International Journal of Computer Vision, vol. 104, 2013.
[35]K. Simonyan and A. Zisserman, “Very Deep Convolutional Networks for Large-Scale Image Recognition,” arXiv:1409.1556 [cs], Sep. 2014.
[36]C. Szegedy et al., "Going deeper with convolutions," in Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 1-9, 2015.
[37]"TensorFlow," [Online]. Available: https://www.tensorflow.org/.
[38]J. Redmon et al., "You only look once: Unified, real-time object detection," in Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 779-788, 2016.
[39]R. Girshick et al., "Rich feature hierarchies for accurate object detection and semantic segmentation," in Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 580-587, 2014.
[40]R. Girshick, "Fast r-cnn," in Proceedings of the IEEE international conference on computer vision, pp. 1440-1448, 2015.
[41]S. Ren et al., "Faster r-cnn: Towards real-time object detection with region proposal networks," in Advances in neural information processing systems, pp. 91-99, 2015.
[42]J. Redmon and A. Farhadi, "YOLO9000: better, faster, stronger. ," CoRR abs/1612.08242, 2016.
[43]A. Neubeck and L. V. Gool, "Efficient Non-Maximum Suppression," in 18th International Conference on Pattern Recognition (ICPR'06), vol. 3, pp. 850-855, doi: 10.1109/ICPR.2006.479, 2006.
[44]J. Redmon and A. Farhadi, "Yolov3: An incremental improvement," arXiv preprint arXiv:1804.02767, 2018.
[45]A Closer Look at YOLOv3. [Online]. Available:
https://www.cyberailab.com/home/a-closer-look-at-yolov3. [Accessed: 12 - Jun - 2019].
[46]Liu, Wei et al. "SSD: Single Shot Multibox Detector". Computer Vision – ECCV 2016 (2016): 21-37.