臺灣博碩士論文加值系統

印刷電路板(Printed Circuit Board; PCB)為目前電子資訊產品之重要零件，其表面包含許多相關SMD製程或技術的電子零件。其長期易遭受外在的濕氣與污塵，甚至陽光的侵蝕，進而導致焊點間容易造成短路老化等現象，致使產品的壽命減短。如今防水膠主要應用於PCB版上的絕緣，防濕、防水及保護作用提供了延長產品壽命的功能。由於在噴膠檢測過程中，因機械手臂噴口力道和距離造成溢膠、缺膠和滴漏的現象而導致載板沾滿不均勻的防水膠，進而使得背景複雜化，單純運用影像分割技術無法準確分割PCB版;這裡透過BPN(Back Propagation Neural Network) 倒傳遞類神經網路做於PCB板與背景作為分類器為區分複雜背景的方法。
由於在檢測區塊裡因SMD相關電子零件的顏色造成檢測誤判，本研究曾透過影像處理的特徵匹配方法，只能達致40%的萃取效益；本研究討論透過SegNet的深度學習的語意分割方法，在測試資料可達至98%的準確率和疊合率MeanIOU為52%，並且探討在有限設備下分類最佳的MeanIOU的評分標準。

The printed circuit board (PCB) is currently an important part of electronic information products, containing many electronic components that related to SMD processes . It suffers external moisture ,dust, and even the erosion of the sun in long time, so which causes to occur between solder joints, resulting in shortened product life.
Currently, waterproof glue which provides the function of extending product life is mainly used in insulation of PCB, moisture proof, waterproof and protective functions .
The carrier board is covered with a non-uniform waterproof glue due to the situation of overflow, lack of glue, and dripping .Which is caused by the force and distance of the robotic arm in spraying. It is not used for separating PCB from complicated background to apply Digital image process method. However, Back Propagation Neural Network (BPN) is used to classify PCB and backgrounds as a method to distinguish complex backgrounds.
Because SMD component’s color caused misjudgment, we had used feature matching method of digital image processing to find detection, but it just achieved 40% performance. In this paper ,we are not only going to research using the SegNet segmentation of deep learning to achieve 98% performance as well as meanIOU 52% ,but discussing every MeanIOU in the limitable equipment.

摘 要 i
ABSTRACT ii
誌 謝 iii
目錄 iv
圖目錄 vi
表目錄 viii
第一章 緒論 1
1.1研究背景與動機 1
1.2研究目的 2
1.3研究範圍與限制 2
1.4研究架構流程 2
第二章 文獻探討 4
2.1 Gerber 4
2.2 人工神經網絡 4
2.3多層感知器（MLP） 5
2.4倒傳遞類神經網路 5
2.5 激活函數 6
2.6圖像語意分割相關文獻 8
2.6.1 Convolutional Neural Networks work 8
2.6.2 Fully Convolutional Networks 9
2.6.3 VGGNet 11
2.6.4 SegNet 11
2.7 SegNet架構 13
2.7.1 Local Response Normalization(局部響應標準化) 14
2.7.2批量標準化 (batch Normalization;BN) 16
2.7.3 ReLU(Rectified Linear Units)激活函數 20
2.7.4 權重正交化 21
2.7.5 MSRA初始化 25
2.7.6 Mean Intersection over Union (MeanIoU) 26
第三章 研究方法 27
3.1 概述 27
3.2 硬體機構設置 29
3.3 分割PCB板與背景 29
3.4 影像資料處理 30
3.4.1資料前處理(Labelme) 30
3.5 SegNet參數設置 32
3.5.1批量大小(Batch size) 32
3.5.2 捲積化 32
3.5.3反捲積 34
3.6 MeanIOU度量 36
第四章 實驗結果與分析 37
4.1 PCB板與背景分割 37
4.2 應用SegNet分割SMD零件 38
4.2.1數據資料 39
4.3實驗結果分析 39
4.3.1權重初始化分析 39
4.3.2 訓練分析結果 42
第五章 結論與未來研究方向 48
5.1 討論 48
5.2 建議 49
5.3 未來研究方向 49
參考文獻 50

1.Jonathan Long,Evan Shelhamer,Trevor Darrell,UC Berkeley. "Fully Convolutional Networks for Semantic Segmentation".
2.Krizhevsky, I. Sutskever, and G. E. Hinton. "Imagenet classification with deep convolutional neural networks". In NIPS, 2012.
3.K. Simonyan and A. Zisserman.Very. "deep convolu-tional networks for large-scale image recognition". CoRR, abs/1409.1556, 2014.
4.C. Szegedy, W. Liu, Y. Jia, P. Sermanet, S. Reed,D. Anguelov, D. Erhan, V. Vanhoucke, and A. Rabinovich. "Going deeper with convolutions".
5."A deep convolutional encoder-decoder architecture for image segmentation". V Badrinarayanan, A Kendall, R Cipolla.
6.Ioffe S, Szegedy C. " Batch normalization: Accelerating deep network training by reducing internal covariate shift ". arXiv:1502.03167, 2015.
7."Deep Residual Learning for Image Recognition". Kaiming He Xiangyu Zhang Shaoqing Ren Jian Sun Microsoft Research.
8."Semantic Image Segmentation with Deep Convolutional Nets and Fully Connected CRFs".Liang-Chieh Chen, George Papandreou, Iasonas Kokkinos, Kevin Murphy, Alan L. Yuille(Submitted on 22 Dec 2014 (v1), last revised 7 Jun 2016 (this version, v4))
9.Z. Liu, X. Li, P. Luo, C. C. Loy, and X. Tang. "Semantic image segmentation via deep parsing network". In ICCV, 2015.
10.Dugas, C.,Y. Bengio, F. Belisle, C. Nadeau and R. Garcia (2001) Incorporating Second-Order Functional Knowledge for Better Option Pricing, "Advances in Neural Information Processing Systems".
11.Y. Gal and Z. Ghahramani. "Bayesian convolutional neural networks with bernoulli approximate variational inference". arXiv:1506.02158,2015
12."Imagenet classification with deep convolutional neural networks A Krizhevsky", I Sutskever, GE Hinton Advances in neural information processing systems, 1097-1105
13."Exact solutions to the nonlinear dynamics of learning in deep linear neural networks"
AM Saxe, JL McClelland, S GanguliarXiv preprint arXiv:1312.6120
14.K. Jarrett, K. Kavukcuoglu, M. A. Ranzato, and Y. LeCun. "What is the best multi-stage architecture for object recognition? " In International Conference on Computer Vision, pages 2146–2153. IEEE, 2009.
15.https://baike.baidu.com/it*em/Gerber%E6%96%87%E4%BB%B6
16.Kirk Baker, Singular Value Decomposition Tutorial, 2005.
17.Y. Gal and Z. Ghahramani. "Bayesian convolutional neural networks with bernoulli approximate variational inference". arXiv:1506.02158,2015.
18."A Review on Deep Learning Techniques Applied to Semantic Segmentation". Alberto Garcia-Garcia, Sergio Orts-Escolano, Sergiu Oprea, Victor Villena-Martinez, Jose Garcia-Rodriguez.Submitted to TPAMI on Apr. 22, 2017
19.https://github.com/wkentaro/labelme
20.JackBouvrie. " Notes on Convolutional Neural Networks". 2006
21.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.
22.Oludolapo Akanni Olanrewaju, Charles Mbohwa, " Comparison of Artificial Intelligence Techniques for Energy Consumption Estimation", 16th annual IEEE Electrical Power and Energy Conference (EPEC), 2016, Ottawa, Canada; 12-14 October 2016
23."Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet" ClassificationK He, X Zhang, S Ren, J Sun International Conference on Computer Vision (ICCV), 2015
24.Y. Gal and Z. Ghahramani. "Bayesian convolutional neural networks with bernoulli approximate variational inference". arXiv:1506.02158,2015.
25."Rethinking atrous convolution for semantic image segmentation". LC Chen, G Papandreou, F Schroff, H Adam arXiv preprint arXiv:1706.05587,2017
26.Gerber File Format Specification.pdf. Ucamco. February 2013 [21 December 2012].
27.New Gerber Format Specification free at http://www.ucamco.com/cn/guest/downloads.
Ucamco Offers Latest Gerber Format Specification.Ucamco.com. February 19, 2013 [February 15, 2013].
28.Karel Tavernier. Improving CAD to CAM Data Transfer: A Practical Approach (PDF). Journal of the HKPCA. 2011/2Q [2 October 2011].
29.http://www.cnblogs.com/52machinelearning/p/5821591.html
30.https://github.com/tkuanlun350/Tensorflow-SegNet
31.https://github.com/vdumoulin/conv_arithmetic