臺灣博碩士論文加值系統

隨著電子產業的蓬勃發展，同時也擴大了零組件業的需求。印刷電路板 (Printed Circuit Board；PCB) 是電子產業中重要的元件，而銅箔基板 (Copper Clad Laminate；CCL) 則是構成印刷電路板的主要材料。
目前工業製程中，銅箔基板的品質檢測主要還是依靠品管人員以目測的方式來進行檢測。但檢測工作是一項高重覆性、枯燥乏味的工作，以人工來檢測雖然可改善品質，但是這不僅耗時沒效率且不符合經濟成本，並容易造成人眼的疲勞。加上其判別依據取決於個人的經驗，沒有一定的準則，容易導致人為的誤差。
如果有效的利用電腦視覺技術，則可以避免人為疏失、降低成本、增加效率及減少傷害發生。本研究探討有關電腦視覺技術在銅箔基板缺陷上的自動檢測。利用小波轉換 (Wavelet Transformation；WT) 與反小波轉換 (Inverse Wavelet Transform；IWT) 找出影像中的缺陷，經轉換後使用區塊分析 (Blob analysis) 找出缺陷中 20 個特徵值，將區塊分析後的特徵值透過約略集合理論 (Rough Set Theory；RST) 進行特徵值的篩選，經篩選過後的特徵值再利用支向機 (Support Vector Machines；SVM) 執行缺陷分類的動作。
本研究在最後缺陷分類過程中，比較 SVM 與 BP 此二種分類方法，所得之結果顯示，SVM 以核心函數為 RBF 之分類結果較好。在各種特徵值組合中，其平均訓練與測試的正確率 (Accuracy) 達到 88.525 % 與 82.435 %，而 BP 的平均正確率為 82.663 % 與 81.93 %。透過分類的正確率可以得知，SVM 應用在銅箔基板缺陷的分類上比 BP 更為優越。

Along with the prospering electronic industry, it also expands the needs over the Parts & Components Industry. Printed Circuit Board (PCB) is an important component in the electronic industry, and Copper Clad Laminate (CCL) is the main material to form a printed circuit board.
Currently in the industry’s manufacturing processes, the quality inspections of PCB are still dependent upon the quality control personnel who will proceed testing by visual inspection. However, the inspection is a kind of highly repetitive and boring task; although manually inspection can improve quality, it is time-consuming, inefficient, and ignorant of economic costs, as well to easily cause fatigues of human eyes. In addition, its judgement is dependent upon personal experiences without exact standards that can easily lead to personal errors.
If we can efficiently apply the computer vision technologies, then we can avoid human errors, cut down costs, increase efficiency, and reduce injuries incidences. This study is mainly investigated the automatic inspection that applied the computer vision technologies to the defects on CCL. First of all, it adopted the Wavelet Transformation (WT) and Inverse Wavelet Transform (IWT) to find the defects in the images. Next, after the process of transformation, it calculated 20 features in the defects by using the blob analysis, and then conducted the screening process of features by mean of the Rough Set Theory (RST). At last, it carried out the defects classifying process on the screened features by using the Support Vector Machines (SVM).
In the final process of defects classification, this study compared both SVM and BP classifying methods, and the results indicated that when the kernel function of SVM is RBF, it will has better classifying outcomes of accuracy. In the various features combinations from SVM, the average accuracy rates of trainings and testings are achieved 88.525% and 82.435% respectively, while those of the average accuracy rates of BP are reached 82.663% and 81.93% respectively. Understanding by the accuracy rates of classification, SVM applications are more extraordinary than BP on the classification of defects for CCL.

中文摘要 i
英文摘要 ii
誌謝 iv
目錄 v
表目錄 vii
圖目錄 viii
第一章 緒論 1
1.1 前言 1
1.2 研究動機 2
1.3 研究目的 2
1.4 研究之範疇與限制 3
1.5 研究架構與流程 3
第二章 文獻探討 5
2.1 銅箔基板 5
2.1.1 空間域法 5
2.1.2 頻譜域法 5
2.2 傅立葉轉換 6
2.3 小波轉換 7
2.3.1 小波轉換四大步驟 9
2.3.2 二維小波 10
2.3.3 反小波轉換 12
2.3.4 小波轉換應用於銅箔基板之缺陷檢測探討 12
2.4 約略集合 13
2.5 支向機 14
第三章 研究方法 18
3.1 視覺系統之基本架構與實驗設備 19
3.2 實驗流程 20
3.3 影像處理 21
3.3.1 平滑 (Smooth) 21
3.3.2 邊緣檢測 23
3.3.3 二值化 24
3.4 二維小波轉換（2D Wavelet Transform） 25
3.5 反小波轉換（Inverse Wavelet Transform） 28
3.6 區塊分析 (Blob analysis) 29
3.7 約略集合 (Rough Set Theory) 30
3.7.1 資訊系統 (Information System) 30
3.7.2 不可區分關聯性 (Indiscernibility Relation) 31
3.7.3 近似集 (Approximation of Sets) 32
3.7.4 屬性篩選與核心 (Core and Reduct of attributes) 33
3.8 支向機 (Support Vector Machine；SVM) 分類 35
3.8.1 線性可分支向機 35
3.8.2 非線性可分支向機 37
3.8.3 Kernel 函數 38
3.9 多類支向機 39
3.9.1 Holdout Method 39
3.9.2 Cross-Validation Method 39
3.9.3 One-against-all method 40
3.9.4 One-against-one method 41
第四章 實驗結果與分析 42
4.1個案簡介 42
4.2銅箔基板類別 44
4.3 銅箔基板之影像處理 45
4.3.1 影像前處理 45
4.3.2 小波轉換與反小波轉換 45
4.3.3 反小波轉換後之影像處理 46
4.4 特徵值擷取與篩選 47
4.4.1 區塊分析 (Blob analysis) 48
4.4.2 約略集合理論 (Rough Set Theory) 49
4.5 SVM分類結果 50
4.5.1 Holdout 分類結果 50
4.5.2 Cross-Validation 分類結果 53
4.5.3 Holdout 與 Cross-Validation 分類結果之比較 55
4.6 比較SVM與BP分類之結果 56
第五章 結論與後續建議 58
5.1 結論 58
5.2後續建議 59
參考文獻 60
附錄 20個特徵值數據 63
作者簡介 92

1. 吳漢銘，降維法在醫學影像分割及微陣列資料分類上的統計應用，國立交通大學統計所，民國92年。
2. 邱學源，導光板品質自動檢測系統之研製，國立高雄第一科技大學碩士論文,機械與自動化工程系，民國93年。
3. 洪永慶，以FPGA實現離散小波轉換並應用於影音壓縮，國立成功大學工程科學研究所，民國92年。
4. 高弋翔，應用支向機於銅箔基板缺陷分類之研究，明新科技大學工程管理研究所碩士論文，民國95年。
5. 郭振鵬，以系統單晶片做印刷電路板銅箔缺陷檢測系統之設計，國立成功大學電機工程研究所，民國93年。
6. 陳雅慧，以基因演算法為基礎建立網頁自動分類機制，中華大學資訊工程系碩士論文，民國92年。
7. 陳豪宇，以元件為基礎之人臉辨識，國立交通大學資訊工程系碩士論文，民國91年。
8. 黃安橦，應用支向機於晶圓圖分類之研究，明新科技大學工程管理研究所碩士論文，民國94年。
9. 馮芝瑋，以支援向量機為基礎之三維臉型識別，長庚大學資訊工程研究所碩士論文，民國91年。
10. 劉雅光，乳癌超音波影像電腦輔助診斷之研究，東海大學訊工程與科學系碩士論文，民國93年。
11. 謝東宏，運用支向機於華語單音節混淆音組辨認之初步研究，長庚大學電機工程研究所碩士論文，民國91年。
12. 繆紹綱 (編著)，數位影像處理－(活用Matlab)，全華科技圖書股份有限公司，民國93年。
13. 繆紹綱 (編著)，數位影像處理，高立圖書股份有限公司，民國92年。
14. 韓歆儀，應用兩階段分類法提昇SVM法之分類準確率，國立成功大學工業管理科學系碩士論文，民國93年。
15. 蘇國瑞，一種使用正交小波表示之影像檢索方法，國立中正大學資訊工程研究所，民國92年。
16. Amet, A. L., A. Ertuzun, and A. Ercil, “Texture defect detection using subband domain co-occurrence matrices”, Image Anal. Interpretation Vol. 1, 1998, pp. 205–210.
17. Arivazhagan, S. and L. Ganesan, “Texture segmentation using wavelet transform”, Pattern Recognition Letters Vol. 24, 2003, pp. 3197-3203.
18. Bileschi, S. M., and Heisele, Bernd., “Advances in component-based face detection”, Pattern Recognition with Support Vector Machines : first International Workshop, 2002, pp. 135-143.
19. Boser, B. E., I. M. Guyon, and V. Vapnik, “A training algorithm for optimal margin classifiers”, In Fifth Annual Workshop on Computational Learning Theory, Pittsburgh, 1992.
20. Cao, L. J. and F. E. H. Tay, “Support vector machine with adaptive parameters in financial time series forecasting”, IEEE Transactions on Neural Networks, Vol. 14, No. 6, 2003, pp. 1506 -1518.
21. Conners, R. W., C.W. McMillin, K. Lin, and R.E. Vasquez-Espinosa, “Identifying and locating surface defects in wood”, IEEE Transactions on Pattern Analysis and Machine Intelligence PAMI Vol. 5, 1983, pp. 573–583.
22. David, A. and B. Lerner, “Pattern classification using a support vector machine for genetic disease diagnosis”, Electrical and Electronics Engineers in Israel, 23rd IEEE Convention of Proceedings, 2004, pp. 289-292.
23. Fletcher, R., “Practical Methods of Optimization”, John Wiley and Sons, Inc, 2nd edition, 1987.
24. Friedman, J., “Another approach to polychotomous classification”, Technical report, Department of Statistics, Stanford University, 1996.
25. Grossmann, A. and J. Morlet, “Decomposition of Hardy functions into square integrable wavelets of constant shape”, SIAM J. Math. Vol. 15, 1984, pp. 723-736.
26. Haralick, R. M., K. Shanmugam, and I. Dinstein, “Textural features for image classification”, IEEE Transactions on Systems, Man and Cybernetics Vol. 3, 1973, pp. 610–621.
27. Lambert, G., and F. Bock, “Wavelet method for texture defect detection, IEEE Int. Conf. Image Process”, Santa Barbara, CA 3, 1997, pp. 201–204.
28. Lemarie, P.G., Y. Meyer, and Ondelettes et “bases Hilbertiennes”, Rev. Mat. Ibero Americana Vol. 2, 1986, pp. 1-18.
29. Liu, S. S. and M.E. Jernigan, “Texture analysis and discrimination in additive noise”, Computer Vision, Graphics and Image Processing Vol. 49, 1990, pp. 52–67.
30. Naqa, I., Yongyi, Yang., M. N. Wernick, N. P. Galatsanos, and R. M. Nishikawa, “A support vector machine approach for detection of microcalcifications”, IEEE Transactions on Medical Imaging, Vol. 21, No. 12 , 2002, pp. 1552 -1563.
31. Nello, C. J., and T. Shave, “An introduction to Support Vector Machines and other kernel-based learning methods”, Cambridge university press, 2000.
32. Pichler, O., A. Teuner, and B. J. Hosticka, “A comparison of texture feature extraction using adaptive Gabor ltering, pyramidal and tree structured wavelet transforms”, Pattern Recognition Vol. 29, 1996, pp. 733-742.
33. Ramana, K.V. and B. Ramamoorthy, “Statistical methods to compare the texture features of machined surfaces”, Pattern Recognition Vol. 29, 1996, pp. 1447–1459.
34. Roman W. Swiniarski and Andrzej Skowron., “Rough set methods in feature selection and recognition”, Pattern Recognition Letters, Vol. 24, 2003, pp. 833-849.
35. Roman W. Swiniarski and Larry Hargis., “Rough sets as a front end of neural-networks texture classifiers”, Neurocomputing, Vol. 36, 2001, pp. 85-102.
36. Shen, Qiang and Richard Jensen., “Selecting informative features with fuzzy-rough sets and its application for complex systems monitoring”, Pattern Recognition, Vol. 37, 2004, pp. 1351-1363.
37. Siew, L. H. and R.M. Hogdson, “Texture measures for carpet wear assessment”, IEEE Transactions on Pattern Analysis and Machine Intelligence Vol. 10, 1988, pp. 92–105.
38. Tsai, D. M. and C. Y. Hsieh, “Automated surface inspection for directional textures”, Image and Vision Computing, Vol. 18, 1999 pp. 49-62.
39. Tsai, Du-Ming and Tse-Yun Huang, “Automated surface inspection for statistical textures”, Image and Vision Computing, Vol. 21, 2003, pp. 307-323.
40. Vapnik, V., “Statistical Learning Theory”, Wiley, 1998.
41. Walczak, B. and D. L. Massart, “Tutorial Rough Sets Theory”, Chemometricsand Intelligent Laboratory Systems, Vol. 47, 1999, pp.1-16.
42. Zeng, Xiangyang and Yanmei Zhan., “Development of a noise sources classification system based on new method for feature selection” Applied Acoustics, Vol. 66, 2005, pp. 1196-1205.