隱形眼鏡生產是一項大量生產工業，由於美觀及消費水平的提升和衛生習慣的進
步，隱形眼鏡 在全球 的需求量逐年提升 ，其製造過程中難免會有瑕疵的產生，傳統的 隱
形眼鏡 瑕疵檢測是以人工目視篩選 或單 一 相機 鏡頭 組 合之 自動化光學檢測 。 這些方法
的缺點是 速度慢及 穩定性差 和 無較佳 的檢測精度 。
本研究設計一套隱形眼鏡檢測系統，
使 用工業相機、倍鏡、鏡頭搭配出兩套
FOV(Field of View)共四個組合 (FOV-1一組及 FOV-2三組 )及 特殊 光源模式 增加檢測時
的精度及 較難檢測之 透明瑕疵的檢出。 隱形眼鏡主要的瑕疵中包含破 損 、刮痕、附加物、
毛邊、氣泡、 P點 (未完全穿透之氣泡 )等。本論文發展出特殊 的 瑕疵 取像 架構 可讓本套
檢測系統更容易檢出附加物、氣泡 瑕疵 以及 刮痕和 P點 透明 瑕疵。
本論文
使用 的 四組 相機鏡頭組合 皆搭配 自動變焦模組 ，解決隱形眼鏡高低落差之對
焦問題 。 擷取隱形眼鏡影像後，利用 YOLO神經網路架構對影像進行神經網路訓練 ，再
進行瑕疵檢測取得結果 。本文亦將 說明此四組 相機鏡頭組合 分工 取像 的 結果 優勢 以 驗
證本系統之 有效性 及實用性。

Contact lens production is a mass production industry. Due to the improvement of aesthetics, buying power, and the increased hygiene standard, the global demand for contact lenses is increasing year by year. It is inevitable that defects will occur on contact lenses during the manufacturing process. Traditional defect detections comprise automatic optical inspection with a single camera lens and labor-intense manual visual inspection. Disadvantages of manual inspection are slow and poor consistence, while for AOI with single camera the limited accuracy
This research proposes a set of contact lens detection system, comprises industrial cameras, magnifiers, and lenses to match two sets of FOV (Field of View), a total of four combinations (FOV-1 group and FOV-2 three groups) and special light source modes. This unique system can increase the detection accuracy and the detection capabilities on transparent defects. The main defects of contact lenses include cracks, scratches, appendages, burrs, bubbles, P-points (bubbles that are not completely penetrated), etc. This thesis develops a special defect imaging framework, which makes it easier for this inspection system to detect additions, bubbles, scratches, and P-point transparent defects.
The four camera-lens combinations used in this thesis are all equipped with an automatic zoom module to solve the focus problem due to contact-lens height difference. After capturing the contact lens image, the YOLO neural network is used to perform defect detections. This thesis will demonstrate the effectiveness and practicability of the system by showing the advantages of the proposed system on a number of experiments.

[1]王滿堂, "隱形眼鏡學上、下冊," 2005. 藝軒圖書出版社.
[2]康茂龍, "機器視覺檢測系統攝影機的應用," p. 第三十卷第三期.
[3]Euresys, "線掃描採集原理."
[4]BlogMech, "Area scan vs Line scan."
[5]新亞洲儀器股份有限公司, "機器視覺『鏡頭』Q&A,"
[6]康茂龍, "視覺檢測系統光源之應用," p. 第二十九卷第四期.
[7]新亞洲儀器股份有限公司, "機器視覺『相機』Q&A,"
[8]Itread.com, "機器視覺，光源筆記," pp.
[9]ADLINK, "影像擷取卡 PCIe-U300 系列."
[10]OPTOTUNE, "可調焦液態鏡頭."
[11]A. Bochkovskiy, C.-Y. Wang, and H.-Y. M. Liao, "Yolov4: Optimal speed and accuracy of object detection," arXiv preprint arXiv:2004.10934, 2020.
[12]ultralytic, "yolov5,"
[13]江大白, "Yolov3&Yolov4&Yolov5模型權重及網絡結構圖."
[14]R. Xu, H. Lin, K. Lu, L. Cao, and Y. Liu, "A Forest Fire Detection System Based on Ensemble Learning," Forests, vol. 12, no. 2, p. 217, 2021.
[15]K. He, X. Zhang, S. Ren, and J. Sun, "Spatial pyramid pooling in deep convolutional networks for visual recognition," IEEE transactions on pattern analysis and machine intelligence, vol. 37, no. 9, pp. 1904-1916, 2015.
[16]S. Liu, L. Qi, H. Qin, J. Shi, and J. Jia, "Path aggregation network for instance segmentation," in Proceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 8759-8768.
[17]H. Jung et al., "Detection of masses in mammograms using a one-stage object detector based on a deep convolutional neural network," PloS one, vol. 13, no. 9, p. e0203355, 2018.
[18]D. Zhou et al., "Iou loss for 2d/3d object detection," in 2019 International Conference on 3D Vision (3DV), 2019: IEEE, pp. 85-94.
[19]H. Rezatofighi, N. Tsoi, J. Gwak, A. Sadeghian, I. Reid, and S. Savarese, "Generalized intersection over union: A metric and a loss for bounding box regression," in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019, pp. 658-666.
[20]偲倢科技, "工業相機規格解釋-視野(FOV)."
[21]E. Optics®, "鏡頭墊圈，墊片和焦距擴展器，像距及物距（分別為I和O）與鏡頭焦距 (f)之 間關係的圖解."