自動化視覺缺陷檢測在科技業佔著重要的角色。發光二極體(LED)具有壽命長、節能和耐用等優點，目前台灣科技業各廠所針對不同的製造過程而製作出的LED透鏡檢測，多以大量的人力目視方式進行檢測，易因視覺疲勞而造成誤判，準確性大大降低。本論文主要以整個高功率發光二極體，針對透鏡的缺陷做檢測。透鏡缺陷分為四種類型：包括刮痕、粒子、變形和氣泡。首先，使用外部環狀光源照明，然後利用CCD擷取影像。在影像處理過程中，先利用中值濾波器降低脈衝雜訊，再經由影像相減擷取缺陷區塊，配合形態學中的補洞運算，使缺陷輪廓完整。在特徵值方面，選擇真圓度、厚度、質心與邊緣灰階比值。在樣本方面，搜集70筆缺陷樣本，而分類器方面使用倒傳遞類神經網路與模糊類神經網路。本實驗進行，結果顯示在訓練樣本為20筆時，測試樣本為50筆時，兩者分類器其辨識率皆可達到100%。然後在透鏡未來製程中可能會有無法預期的突發情況，所以再設計另一組測試樣本，經由分類器去訓練和測試，最後驗證了模糊類神經網路比倒傳遞類神經網路有更好的強健性，可獲得相當準確的辨識率，成功被應用於發光二極體透鏡缺陷自動檢測系統。
關鍵詞:發光二極體、影像處理、分類器、倒傳遞類神經網路、模糊類神經網路。

Automatic visual inspection of defects plays an important role in technology industry. Currently, the light emitting doide (LED) has advantages of a long life, energy saving, and durability. Generally, its lens inspection is carried out by human eyes. Visual fatigue easily causes misjudgement; thus greatly reducing accuracy. This paper aims to apply image processing and classifiers to inspect lens defects of the high-power LED. Lens defects include scratch, particle, deformation and bubble. First, we use external ring light and a charge coupled device (CCD) to capture image. In image processing, the median filter is uesd to reduce impulse noise, and then image subtraction gives the defect region, and the hole filling in morphological operation obtains the complete shape of the defect. We choose roundness, thickness, shading as defect features. Seventy defect samples are collected, and the classifiers of back-propagation neural network and fuzzy neural network are used. The experiment result shows that with twenty training samples and fifty testing samples, the two classifiers can have classification rates of 100%. With some artificially generated feature values in testing samples, the experiment shows that the fuzzy neural network is more robust than the back-propagation neural network and has higher classification rate. Thus, the automatic defect inspection system works well for the high-power LED.

[1] G. S. Cho and Y. S. Kwon, “AlGaAs/GaAs lens shaped LED with high efficiency and narrow field pattern,” The International Society for Optical Engineering, Vol. 1813, No. 12, pp. 142-153, 1992.
[2] C. Y. Chang, C. H. Li, S. Y. Lin and M. D. Jeng, “Application of two hopfield neural networks for automatic four element LED inspec -tion,” IEEE Trans. on System, Man and Cybernetics, Vol. 39, No. 3, pp. 352-365, 2009.
[3] H. D. Lin, “Automated defect inspection of light-emitting diode chips usin gneural network and statistical approaches,” Expert Systems with Applications, Vol. 36, pp. 219-226, 2009.
[4] Y. Luo, Z. He and Y. Zhu, “Led’s efficiency-current characteristics of automatic inspection system,” Optical Technique, Vol. 37, No. 3, pp. 336-340, 2011.
[5] H. Y. Tan and T. W. Ng, “Light-emitting-diode inspection using a flatbed scanner,” Optical Engineering, Vol. 47, No. 47, pp. 103602, 2008.
[6] W. C. Chen and S. W. Hsu, “A neural-network approach for an automatic LED inspection system,” Expert Systems with Applications, Vol. 33, No. 2, pp. 531-537, 2007
[7] J. C. Shyu, K. W. Hsu, K.S. Yang and C. C. Wang, “Performance of the LED array panel in a confined enclosure,” 2010 5th International Microsystems Packaging Assembly and Circuits Technology Conference, Vol. 20 pp. 1-4, 2010.
[8] D. P. Mital and E. K. Teoh, “Computer based wafer inspection system,” Proceeding of International Conference on Industrial Electronics, Vol. 3, pp. 2497-2503, 1991.
[9] W. Y. Wu, M. J. Wang and C. M., Liu, “Automated inspection of printed circuit boards through machine vision,” Computers in Industry, Vol. 28, No. 2, pp. 103-111, 1996.
[10] C. I. Christodoulou, S. C. Michaelides and C. S. Pattichis, “Multifeature texture analysis for the classification of clouds in satellite imagery,” IEEE Trans. on Geoscience and Remote Sensing, Vol. 41, No. 11, pp. 2662-2668, 2003.
[11] C. D. Ruberto, G. Rodriguez and S. Vitulano, “Image segmentation by
texture analysis,” 10th International Conference on Image Analysis
and Processing, pp. 376-381, 1999.
[12] M. Sezgin and B. Sankur, “Survey over image thresholding techniques and quantitative performance evaluation,” Journalof Electronic Imaging, Vol. 13, pp. 146-165, 2004.
[13] C. T. Su, T. Yang and C. M. Ke, “Neural network approach for semiconductor wafer post sawing inspection,” IEEE Trans. on Semiconductor Manufacturing, Vol. 15, No. 2, pp. 260-266, 2002.
[14] 林柏羽，應用倒傳遞網路於電子產品之LED自動化檢測系統，大同大學資訊工程研究所，2005。
[15] 蔡孟凱編著，C++ Builder6完全攻略，上奇科技股份有限公司， 2003。
[16] 余明興編著，Borland C++ Builder程式設計精典，文魁資訊股份有限公司，2002。
[17] 陳隆建，發光二極體之原理與製程，全華科技，2008。
[18] N. R. Bonda, K. Kaskoun, W. H. Lytle, B. Marlin, G. Swan, J. W. Stafford, and G. Tam, “Physical design and assembly process development of a multichip package containing a light emitting diode array die,” IEEE Trans. on Components, Vol. 20, No. 4, pp. 389-395, 1997.
[19] Y. C. Hsu, Y. K. Lin, M. H. Chen, C. C. Tsai, J. H. Kuang, S. B. Huang, H. L. Hu, Y. I. Su and W. H, “Failure mechanisms associated with lens shape of high-power LED modules in aging test,” IEEE Trans. on Electron Devices, Vol. 55, No. 2, pp. 689-694, 2008.
[20] A. Okuno and O. Tanaka, “High bright white LED lens formation by vacuum printing encapsulation systems (VPES) and its packaging resin,” International Conference on Electronic Packaging Technology and High Density Packaging, pp. 448-450, 2009.
[21] 鐘國亮編著，影像處理與電腦視覺第三版，東華書局，2005。
[22] Y. Du, C. I. Chang and C. Y. Thouin, “Unsupervised approach to color video thresholding,” Optical Engineering, Vol. 43, No. 2, pp. 282-289, 2004.
[23] P. B. Chou, A. R. Rao, M. C. Sturzenbecker, F. Y. Wu and V. H. Brecher, “Automatic defect classification for semiconductor manufacturing,” Machine Vision and Applications, Vol. 9, No. 4, pp. 201-214, 1997.
[24] 葉怡成，類神經網路模式應用與實作，儒林圖書有限公司，1995。
[25] 王進德，類神經網路與模糊控制理論入門與應用，全華圖書公司，2007。
[26] L. X. Wang, “A course in fuzzy systems and control,” New Jersey: 122 Prentice-Hall, 1997.
[27] K. S. Narendra and K. Parthasarathy, “Identification and control of dynamical systems using neural networks,” IEEE Trans. Neural Networks, Vol. 1, No. 1, pp. 4-27, 1990.
[28] J. Zhang and A. J. Morris, “Recurrent neural-fuzzy networks for nonlinear process modeling,” IEEE Trans. Neural Networks, Vol. 10, No. 2, pp. 313-326, 1999.