臺灣博碩士論文加值系統

本篇論文利用中國鋼鐵股份有限公司所生產的小鋼胚，進行小鋼胚影像身分識別，傳統的方法在每批生產出來的小鋼胚烙印生產編號，讓工作人員能夠透過編號來識別小鋼胚的身分，但因小鋼胚在生產時表面溫度高達700度至900度，即便烙印生產編號也容易因周遭環境而導致小鋼胚表面出現缺陷損壞的情形，為了改善這樣的情況，在本篇論文中將嘗試以小鋼胚表面紋路影像並搭配影像識別技術，來取代傳統人工識別的方法，進而提升產線生產的效率。
而本篇所使用之技術將以深度學習技術來實作出我們所需要結果，將多張小鋼胚表面紋路影像利用CNN來做訓練，並在最後對圖片進行識別，辨識出該張影像的編號是屬於哪一塊小鋼胚的結果。

This paper proposes a method for image recognition which uses steel billets pro-duced by China Steel Corporation. To be recognize by human easily, the company has to print serial numbers on each steel billets. Steel billets are produced in a high temperature environment. The production may cause damage to these serial numbers. We decide to use automatic image recognition technology to improve the efficiency and precision.
We use deep leaning technology to approach our target. We use images to train the CNN model in order to recognize these images. By using the model, we can identify each steel billets by image recognition.

摘要 i
ABSTRACT ii
致謝 iii
目錄 iv
表目錄 vi
圖目錄 vii
第一章 、簡介 1
1.1 研究動機 1
1.2 研究目的 1
1.3 論文架構 2
第二章 、相關技術探討 3
2.1 神經網路模型 3
2.1.1人腦視覺原理 3
2.1.2類神經網路 4
2.1.3人工神經元模型 4
2.2 卷積神經網路 5
2.2.1卷積神經網路的網路結構 5
2.2.2局部感知 6
2.2.3權重共享 7
2.2.4池化層 7
2.2.4 全域連接 8
2.3 深度學習網路模型-AlexNet 9
2.4 深度學習工具 - Caffe 10
2.4.1 Caffe的特點 10
2.4.2 Caffe的架構 11
第三章 、研究方法 12
3.1 訓練環境 12
3.2 Stirmark轉換圖片 12
3.3 建立資料集 18
3.4 利用AlexNet網路訓練 19
3.5 AlexNet網路運作流程 19
第四章 、研究結果 23
4.1 Accuracy&loss曲線圖 23
4.2 結果比對 24
第五章 、結論 27
參考文獻 28

[1]DAVID G. LOWE, “Distinctive Image Features from Scale-Invariant Keypoints,” International Journal of Computer Vision, vol. 60, pp. 91–110, Nov. 2004.
[2]Herbert Bay, Tinne Tuytelaars, and Luc Van Gool, “SURF: Speeded Up Robust Features,” European Conference on Computer Vision, ECCV, pp. 404-417, 2006.
[3]Ashwani Kumar Dubey and Zainul Abdin Jaffery, “Maximally Stable Extremal Region Marking (MSERM) based Railway Track Surface Defect Sensing,” IEEE Sensors Journal, vol. 16, pp. 9047-9052, Oct. 2016.
[4]Sang Jun Lee, Jaepil Ban, Hyeyeon Choi and Sang Woo Kim,” Localization of slab identification numbers using deep learning”, 2016 16th International Con-ference on Control, Automation and Systems (ICCAS), pp. 16-19, Oct. 2016.
[5]A. Krizhevsky, I. Sutskever, and G. E. Hinton, “ImageNet classification with deep convolutional neural networks.” Neural Information Processing Systems 2012, Dec. 2012.
[6]D. H. Huble, T. H. Wiesel, “Receptive fields of single neuroes in the cat’s striate cortex”, Journal of Physiology, pp. 574-591, Oct. 1959.
[7]WARREN S. McCULLOCH, WALTER PITTS, “A Logical Calculus of the Ide-as Immanent in Nervous Activity,” The Bulletin of Mathematical Biophysics, vol. 5, pp. 115-133, Dec.1943.
[8]F. ROSENBLATT, “The Perceptron: A Probabilistic Model for Information Storage And Organization in The Brain,” .Psychological Review, Vol. 65, No. 6, pp. 386-408, Feb. 1958.
[9]J. J. Hopfield, D. W. Tank, “ “Neural” Computation of Decisions in Optimization Problems,” Biological Cybernetics, vol. 52, pp.141-152, Jul. 1985.

[10]G. E. Hinton, T. J. Sejnowski and D. H. Ackley, “Boltzmann Machines: Constraint Satisfaction Networks that Learn,” Department of Computer Science, CMU-CS-84-119,May. 1984.
[11]Gail A. Carpenter, Stephen Grossberg, and David B. ROSEN, “Fuzzy ART: Fast Stable Learning and Categorization of Analog Patterns by an Adaptive Resonance System,” Neural Networks, vol. 4, pp. 759-771, 1991.
[12]Yoshua Bengio. “Learning deep architectures for AI,” Foundations and Trends in Machine Learning, vol. 2, no. 1, pp. 1-127, Nov. 2009.
[13]D. H. Hubel, T. N. Wiesel, “Receptive Fields, Binocular Interaction and Func-tional Architecture in the Cat’s Visual Cortex,” Journal of Physiology, vol. 160, pp. 106-154, Jan. 1962.
[14]KUNIHIKO FUKUSHIMA and SEI MIYAKE. “Neocognitron: a new algo-rithm for pattern recognition Tolerant of Deformations and Shifts in Position,” Pattern Recognition, vol. 15, no. 6, pp. 455-469, 1982.
[15]D. E. Rumelhart, G. E. Hinton, R. J. Williams, “Learning representations by back-propagating errors,” Letters to Nature, Nature 323, pp. 533-536, Oct. 1986.
[16]Y. Jia, E. Shelhamer, J. Donahue, S. Karayev, J. Long, R. Girshick, S. Guadarrama, and T. Darrell. “Caffe: Convolutional architecture for fast feature embedding,” Proceedings of the 22nd ACM international conference on Mul-timedia, pp. 675-678, Nov. 2014.
[17]T.Dettmers, Which GPU to Get for Deep Learning: My Experience and Advice for Using GPUs in Deep Learning, http://timdettmers.com/2017/04/09/which-gpu-for-deep-learning/.
[18]Yangqing Jia, and Evan Shelhamer, Caffe Tutori-al,http://caffe.berkeleyvision.org/tutorial/, 2016.
[19]Stirmark Benchmark by, http://www.petitcolas.net/watermarking/stirmark/
[20]Christian Szegedy, Wei Liu, Yangqing Jia, Pierre Sermanet, Scott Reed, Dragomir Anguelov, Dumitru Erhan, Vincent Vanhoucke and Andrew Rabinovich, “Going Deeper with Convolutions,” IEEE Conference on Computer Vision and Pattern Recognition (CVPR)