近年來機器學習(Machine Learing)被廣泛應用於製造業中，其中一個重要的議題即為機器視覺(Machine vision)，其目的在於進行產品品質檢測及工安監控，以便即早發現異常進行處置。
然而以面板業G3~4代廠而言，由於設備年限已久，若需改造新購自動檢測設備，則往往因為設備改造或新購成本較高，且無法遍布所有製程站點，產品缺陷攔檢收效甚微，致使公司營運策略趨於保守，至今仍仰賴既有人力抽驗檢測作業的模式繼續運作。
本研究藉由文獻探討及影像建模方式，利用深度學習之卷積神經網路架構(Convolutional Neural Networks，CNN)模型，參照YOLOv3架構並搭配影像預處理自行架設影像辨識系統，實現動態影像辨識之應用於製造現場產品缺陷檢測，藉此強化完善製程站點之品質監控系統，並作為機器視覺應用於生產現場管理之實務研究參考。

In recent years, machine learning (Machine Learing) has been widely used in manufacturing industry. One of the important topics is Machine Vision. The purpose is to perform product quality inspection and industrial safety monitoring, so that abnormalities can be detected and disposed of early.
However, for the G3 ~ 4 generation plants in the panel industry, because the equipment has been in use for a long time, if it is necessary to modify the newly purchased automatic detection equipment, it is often because the equipment modification or new purchase cost is high, and it cannot be spread across all process sites, and product defects are blocked The check and collection effect is very small, which makes the company's operating strategy tend to be conservative. So far, it still relies on the existing manual sampling and testing operation mode to continue operation.
In this study, by using literature research and image modeling methods, using deep learning's Convolutional Neural Networks (CNN) model, referring to YOLOV3 architecture, and using image pre-processing to self-set up an image recognition system to achieve dynamic images Identification is used to detect product defects at the manufacturing site, thereby strengthening and improving the quality monitoring system at the process site, and as a research reference for machine vision applied to the production site management practice.

中文文獻
[1] 王輝、孫洪(2017)。魯棒主成分分析的鋁箔表面缺陷檢測方法。信號處理雜誌， 33(4)，577-582。
[2] 李力、王耀南、陳鐵健(2016)。大尺寸LCD玻璃基板多視覺缺陷檢測系統研究。控制工程，2，222-226。
[3] 李彥冬、郝宗波、雷航(2016)。卷積神經網絡研究綜述。計算機應用，36(9)， 2508-2515。
[4] 林暉凱(2018)。扣件業盤元價格預測(碩士論文)。國立成功大學，台南市。
[5] 林大貴(2017)。TensorFlow+Keras深度學習人工智慧實務應用。新北市：博碩文化股份有限公司。
[6] 馬政(2019)。基於深度學習與影像處理之儀表偵測與讀數識別(碩士論文)。國立中央大學，桃園市。
[7] 張迪、樊紹勝(2019)。基於YOLO V3的輸電線路故障檢測方法。自動化技術與應用，(7)。27。
[8] 張元翔(2019)。數位影像處理:Python程式實作。新北市:全華圖書。
[9] 湯勃、孔建益、伍世虔(2017)。機器視覺表面缺陷檢測綜述。中國圖像圖形學報(編號：1006-8961(2017)12-1640-24)。
[10] 閔永智、岳彪、馬宏鋒、肖本郁(2018)。基於圖像灰度梯度特徵的鋼軌表面缺陷檢測。儀器儀表學報，39(4)，220-229。
[11] 黃登淵(2019)。基於深度學習與視覺之車輛偵測。科學與工程技術學報，15(2)，1-11。
[12] 楊松儒(2019)。以深度學習為基礎之路面破損與閥栓檢測系統(碩士論文)。臺灣師範大學，台北市。
[13] 楊建華、姚建平、范磊、張端、許心夢(2017)。單板表面裂縫圖像處理方法研究。林業機械與木工設備，45(4)，29-31。
[14] 詹鈞婷(2019)。基於深度學習之心臟血管鈣化偵測系統(碩士論文)。國立中央大學，桃園市。

英文文獻
[1] Blue, S. T., & Brindha, M. (2019). Edge detection based boundary box construction algorithm for improving the precision of object detection in YOLOv3. Paper presented at the 2019 10th International Conference on Computing, Communication and Networking Technologies (ICCCNT).
[2] Bharath, A. A., & Petrou, M. (2008). Next generation artificial vision systems: Reverse engineering the human visual system: Artech House.
[3] Canziani, A., Paszke, A., & Culurciello, E. (2016). An analysis of deep neural network models for practical applications. arXiv preprint arXiv:1605.07678.
[4] Girshick, R. (2015). Fast r-cnn. Paper presented at the Proceedings of the IEEE international conference on computer vision.
[5] Girshick, R., Donahue, J., Darrell, T., & Malik, J. (2014). Rich feature hierarchies for accurate object detection and semantic segmentation. Paper presented at the Proceedings of the IEEE conference on computer vision and pattern recognition.
[6] He, K., Zhang, X., Ren, S., & Sun, J. (2016). Deep residual learning for image recognition. Paper presented at the Proceedings of the IEEE conference on computer vision and pattern recognition.
[7] Kilic, B., Baykal, E., Ekinci, M., Dogan, H., Ercin, M. E., & Ersoz, S. (2019). Automated Nuclei Detection on Pleural Effusion Cytopathology Images using YOLOv3. Paper presented at the 2019 4th International Conference on Computer Science and Engineering (UBMK).
[8] Lee, C.-Y., Chen, C.-H., & Chien, C.-F. (2014). A simulation analysis for evaluating TFT-LCD fab capacity expansion with a distant transportation problem. International Journal of Production Research, 52(6), 1868-1885.
[9] Lee, C.-Y., Huang, T.-S., Liu, M.-K., & Lan, C.-Y. (2019). Data science for vibration heteroscedasticity and predictive maintenance of rotary bearings. Energies, 12(5), 801.
[10] Lee, C.-Y., & Dong, Z.-H. (2019). Hierarchical Equipment Health Index Framework. IEEE Transactions on Semiconductor Manufacturing, 32(3), 267-276.
[11] LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning. nature, 521(7553), 436-444.
[12] Lee, Y., Lee, C., Lee, H.-J., & Kim, J.-S. (2019). Fast Detection of Objects Using a YOLOv3 Network for a Vending Machine. Paper presented at the 2019 IEEE International Conference on Artificial Intelligence Circuits and Systems (AICAS).
[13] Liu, W., Anguelov, D., Erhan, D., Szegedy, C., Reed, S., Fu, C.-Y., & Berg, A. C. (2016). Ssd: Single shot multibox detector. Paper presented at the European conference on computer vision.
[14] Lin, T.-Y., Dollár, P., Girshick, R., He, K., Hariharan, B., & Belongie, S. (2017). Feature pyramid networks for object detection. Paper presented at the Proceedings of the IEEE conference on computer vision and pattern recognition.
[15] Lin, T.-Y., Goyal, P., Girshick, R., He, K., & Dollár, P. (2017). Focal loss for dense object detection. Paper presented at the Proceedings of the IEEE international conference on computer vision.
[16] Mao, Q.-C., Sun, H.-M., Liu, Y.-B., & Jia, R.-S. (2019). Mini-YOLOv3: real-time object detector for embedded applications. IEEE Access, 7, 133529-133538.
[17] Redmon, J., & Farhadi, A. (2019). Yolov3: An incremental improvement. arXiv 2018. arXiv preprint arXiv:1804.02767.
[18] Srivastava, N., Hinton, G., Krizhevsky, A., Sutskever, I., & Salakhutdinov, R. (2014). Dropout: a simple way to prevent neural networks from overfitting. The journal of machine learning research, 15(1), 1929-1958.
[19] Simonyan, K., & Zisserman, A. (2014). Very deep convolutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556.
[20] Szegedy, C., Vanhoucke, V., Ioffe, S., Shlens, J., & Wojna, Z. (2016). Rethinking the inception architecture for computer vision. Paper presented at the Proceedings of the IEEE conference on computer vision and pattern recognition.
[21] Szegedy, C., Ioffe, S., Vanhoucke, V., & Alemi, A. A. (2017). Inception-v4, inception-resnet and the impact of residual connections on learning. Paper presented at the Thirty-first AAAI conference on artificial intelligence.
[22] Yan, Z., Li, B., Chen, W., Young, T., Li, N., & Gao, J. (2010). Automatic inspection system for flat panel substrate. In: Google Patents.
[23] Yanan, S., Hui, Z., Li, L., & Hang, Z. (2018). Rail surface defect detection method based on yolov3 deep learning networks. Paper presented at the 2018 Chinese Automation Congress (CAC).

網頁文獻
[1] Chenchou LO(2017) , YOLO - You Only Look Once介紹, Available from: https:// medium.com/@chenchoulo/yolo-介紹-4307e79524fe,最後瀏覽日:2020/4/11.
[2] Opengate.(2018),Yolo:基於深度學習的物件偵測(含YoloV3),Available from: https://mropengate.blogspot.com/2018/06/yolo-yolov3.html,最後瀏覽日:2020/5/13.
[3] Synioe(2018), YOLOv3深度理解, Available from:https://blog.csdn.net/Synioe/ article/ details/84110504,最後瀏覽日:2020/4/19.
[4] Tommy Huang (2018)。卷積神經網路(Convolutional neural network，CNN) -CNN運算流程。Available from:https://medium.com/@chih.sheng.huang821/卷積神經網路-convolutional-neural-network-cnn-cnn運算流程-ecaec240a631，最後瀏覽日:2020 /5/26。
[5] 每日頭條網站(2019)。機器視覺在工業中的五大典型應用。Available from: https:// kknews.cc/zh-tw/tech/pp9v55j.html，最後瀏覽日:2020/3/08。
[6] 國立台灣大學電子報第0038期(2016)。淺談Deep Learning原理及應用。http:// www.cc.ntu.edu.tw/chinese/epaper/0038/20160920_3805.html，最後瀏覽日：2020/5/26。
[7] 新通訊元件雜誌(2018)。AI/感測關鍵技術助陣-影像辨識應用全面啟動。Available from:https://www.2cm.com.tw/2cm/zh-tw/market/39CF0BA42294406F8EE77337CF6 7FBB5，最後瀏覽日:2020/3/08。
[8] 韓揚銘(2017)。圖像巨量資料時代來臨。Available from: https://smartauto.ctimes.　com.tw/DispArt-tw.asp?O=HK1266CHI5MARASTD3，最後瀏覽日:2020/05/01。