本論文展示一個基於圖像識別之玻璃瓶中液體存在檢測系統的設計與實現。主要工作原理是光在兩種折射率不同的介質之間的界面處折射，以改變前向光的方向。屏幕上特定的紅光圖樣是在玻璃瓶中多次折射光過程後生成的，本設計可以識別玻璃瓶中是否存在液體。
我們採用兩種不同的圖像識別技術去檢測玻璃瓶中的液體存在性。第一種是使用卷積類神經網路訓練的圖像分類，另一種則是使用OpenCV電腦視覺庫的邊緣偵測。紅光雷射源有兩種不同的光圖樣，分別是一字線圖樣及點狀圖樣。
所提出的系統可以在生產線中廣泛實施，以監視帶有液體的玻璃瓶。為了實現本研究的目的，僅需設置一紅光雷射、一具攝影機的樹莓派4 B型開發板、一屏幕和所提出的圖像識別技術即可。

This paper presents the design and implementation of the detection systems of liquid presence in glass bottles based on image recognition. The primary operating principle is the refraction of light at the interface between two media of different refractive indices to change the direction of forwarding light. The specific red light pattern on the screen is generated after many refractive light processes in a glass bottle, which design can give identifying whether liquid presents in glass bottles.
We adopted two different image recognition methods to detect the presence of liquid in the glass bottle, one is the image classification based on convolutional neural network training, and the other is the edge detection based on the OpenCV computer vision library. The red laser source has two kinds of light patterns, such as a line pattern and a dot pattern, respectively.
The proposed systems can widespread implement in the production lines to monitor glass bottles with liquid. To achieve the goal of this research, they are only necessary to set up a red laser, a Raspberry Pi 4 Model B development board with a camera, a screen, and the proposed image recognition techniques.

目錄
誌謝Ⅰ
中文摘要II
ABSTRACTⅢ
目錄IV
圖目錄VII
表目錄VIII
第一章 緒論1
1.1 前言1
1.2 研究目的1
第二章 商用產品探討3
2.1 機器視覺檢測技術簡介3
2.2 基恩斯公司IB系列產品3
2.3 大恒圖像公司DH-FBI-4A滿瓶檢驗機4
2.4 HEUFT公司HEUFT PRIME產品5
第三章 軟體與硬體介紹6
3.1 硬體介紹6
3.1.1 樹莓派4 B型開發板6
3.1.2 攝影機模組6
3.1.3 紅光雷射源7
3.1.4 玻璃瓶9
3.2 軟體介紹10
3.2.1 OpenCV10
3.2.2 Keras11
3.2.3 卷積類神經網路11
3.2.4 Anaconda12
第四章 系統架構設計13
4.1 系統架構簡介13
4.2 系統一13
4.3 系統二19
第五章 實驗結果22
5.1 系統一22
5.1.1 原始圖25
5.1.2 二值化27
5.1.3 空間顏色轉換29
5.1.4 三種圖像處理比較31
5.2 系統二33
第六章 結論與未來展望38
6.1 結論38
6.2 未來展望39
參考文獻40
圖目錄
圖 1偵測試管液位3
圖 2滿瓶檢測4
圖 3液位檢測5
圖 4第四代樹莓派開發板6
圖 5 羅技網路攝影機C930e7
圖 6 一字線紅光雷射源8
圖 7 一字線紅光圖樣8
圖 8 點狀紅光雷射源9
圖 9 點狀紅光圖樣9
圖 10 玻璃瓶10
圖 11 卷積類神經網路架構11
圖 12  Anaconda畫面12
圖 13系統一架構14
圖 14 一字線雷射光路徑圖14
圖 15 圖像訓練流程圖17
圖 16 系統操作流程圖18
圖 17系統二架構19
圖 18點狀雷射光路徑圖20
圖 19系統操作流程圖21
圖 20辨識之有效距離22
圖 21 攝影機加上濾光片比較24
圖 22點狀紅光圖樣25
圖 23 原始圖訓練損失/正確率26
圖 24 原始圖辨識27
圖 25 圖像二值化訓練損失/正確率28
圖 26 二值化辨識29
圖 27 圖像空間顏色轉換訓練損失/正確率30
圖 28 空間顏色轉換辨識31
圖 29 系統一即時辨識實驗設置32
圖 30系統一即時辨識結果33
圖 31 點狀紅光圖樣辨識34
圖 32一字線紅光圖樣35
圖 33 系統二即時辨識實驗設置36
圖 34系統二即時辨識結果37
表目錄
表 1 系統比較39

[1] 研華科技Advantech, “工業自動化,” available link: https://www.
advantech.tw/solutions/ifactory
[2] 台灣基恩斯股份有限公司, “液位檢測,” available link: https://www.
keyence.com.tw/mykeyence/downloadFromDLList?downloadAssetId=AS_94403&downloadZipFlag=0
[3] 大恒圖像, “滿瓶檢測,”  available link: http://www.daheng-imavision
.com/wap/industry-detail.php?infoid=53
[4] HEUFT, “滿瓶檢測,” available link: https://heuft.com/zh/products/
beverage/fill-containers/full-container-check-heuft-prime
[5] Logitech, “rightlight 2,” available link: https://download01.logitech.com/
web/ftp/pub/pdf/cameras/rightLight2_innovation_brief_enu.pdf
[6] 佳皇科技有限公司, “雷射模組,” available link: https://www.led-shop
.com.tw/abc74LD.htm
[7] Anwar, S., Hwang, K., & Sung, W. (2015, April). Fixed point optimization of deep convolutional neural networks for object recognition. In 2015 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) (pp. 1131-1135).
[8] CH. Tseng, “卷積神經網路架構,” available link: https://chtseng
.wordpress.com/2017/09/12/初探卷積神經網路/
[9] Kadiyala, A., & Kumar, A. (2017). Applications of python to evaluate environmental data science problems. Environmental Progress & Sustainable Energy, 36(6), 1580-1586.
[10] Awcock, G. J., & Thomas, R. (1989, April). Binary image sensor developments. In IEE Colloquium on Pattern Recognition for Binary Images (pp. 1-1). IET.
[11] Vitabile, S., Pollaccia, G., Pilato, G., & Sorbello, F. (2001, September). Road signs recognition using a dynamic pixel aggregation technique in the HSV color space. In Proceedings 11th International Conference on Image Analysis and Processing (pp. 572-577). IEEE. 
[12] Chollet, F. (2016). Building powerful image classification models using very little data. Keras Blog. 
[13] Khrisne, D. C., & Suyadnya, I. M. (2018, October). Indonesian Herbs and Spices Recognition using Smaller VGGNet-like Network. In 2018 International Conference on Smart Green Technology in Electrical and Information Systems (ICSGTEIS) (pp. 221-224). IEEE.
[14] Chollet, F. (2018). Introduction to Keras. March 9th.
[15] Van De Ville, D., Nachtegael, M., Van der Weken, D., Kerre, E. E., Philips, W., & Lemahieu, I. (2003). Noise reduction by fuzzy image filtering. IEEE transactions on fuzzy systems, 11(4), 429-436.
[16] Raspberry Pi Blog, “濾光片,” available link: https://www.raspberrypi
.org/blog/whats-that-blue-thing-doing-here/
[17] 知乎, “光斑,” available link: https://zhuanlan.zhihu.com/p/28009549