跳到主要內容

臺灣博碩士論文加值系統

(44.212.96.86) 您好!臺灣時間:2023/12/06 16:29
字體大小: 字級放大   字級縮小   預設字形  
回查詢結果 :::

詳目顯示

我願授權國圖
: 
twitterline
研究生:陳尚謙
研究生(外文):CHEN, SHANG-CHIEN
論文名稱:透鏡之自動化光學瑕疵檢測
論文名稱(外文):Automatic Defect Detection of Optical Lens
指導教授:吳先晃
指導教授(外文):WU, HSIEN-HUANG
口試委員:何前程蔡文凱
口試委員(外文):HO, CHIEN-CHENGTSAI, WEN-KAI
口試日期:2022-07-19
學位類別:碩士
校院名稱:國立雲林科技大學
系所名稱:電機工程系
學門:工程學門
學類:電資工程學類
論文種類:學術論文
論文出版年:2022
畢業學年度:110
語文別:中文
論文頁數:92
中文關鍵詞:光學玻璃透鏡瑕疵檢測自動化光學檢測深度學習
外文關鍵詞:Optical glass lensDefect detectionAutomated optical inspectionDeep learning
相關次數:
  • 被引用被引用:0
  • 點閱點閱:172
  • 評分評分:
  • 下載下載:0
  • 收藏至我的研究室書目清單書目收藏:0
光學玻璃透鏡廣泛應用於各種類型的光學設備、高精密光學元件及高精密鏡頭產品,例如:車載鏡頭、監視器鏡頭、投影機鏡頭、數位相機鏡頭、相關工業用鏡頭、醫療用內視鏡片...等,因使用場域均屬於高科技產品,所以在產品出廠前均須要進行完整的瑕疵檢測,以確保良好的品質。目前業界出廠檢驗,仍依靠人工照射強光並以肉眼仔細檢查鏡片表面來進行外觀瑕疵的檢測。因為檢測過程均依靠人力,常會因疲勞而造成漏檢,且長時間在強光下檢測,也有眼睛受損的風險。而現今產業裡,自動化光學檢測(Automated Optical Inspection, AOI)和人工智慧(Artificial Intelligence, AI)的應用也越來越普及化,許多科技業、傳統產業紛紛導入AOI或AI系統作為主要的產品檢測系統。因此本研究將光學透鏡作為研究之對象,設計一套AI玻璃透鏡檢測系統,使用用不同種類的相機進行拍攝比較,並且用YOLOv4(You Only Look Once)深度學習模型來進行瑕疵檢測,挑選出應用於玻璃透鏡瑕疵檢測效果最好之方法。最終結果,使用工業相機搭配高角度環光,並且使用YOLOv4作為主要深度學習網路來開發系統,同時此玻璃透鏡瑕疵的準確率達到80.7%,並設計一套使用者介面(User Interface, UI),方便使用者更快速地查看光學透鏡上檢測到的瑕疵,使瑕疵資訊更明瞭,以利協助改善產線製程,提升自動化產線效能。
Optical glass lenses are widely used in various types of optical equipment, high-precision optical components and high-precision lens products, such as: car lenses, monitor lenses, projector lenses, digital camera lenses, related industrial lenses, medical endoscopy lenses... etc. Because the use of the field are high-tech products, so the products must be completely tested for defects before shipping to ensure good quality. Because the fields of use are high-tech products, a complete defect inspection must be conducted before the products leave the factory to ensure good quality. At present, the industry still relies on manual irradiation of bright light and careful inspection of the surface of the lens by the naked eye to inspect the appearance of defects. Because the inspection process relies on human labor, it is often caused by fatigue and leakage, and the risk of eye damage due to long hours of inspection under bright light. Nowadays, Automated Optical Inspection (AOI) and Artificial Intelligence (AI) are becoming more and more popular in the industry, and many technology and traditional industries have introduced AOI or AI systems as their main product inspection systems. Therefore, in this study, an AI glass lens inspection system was designed to compare different types of cameras and use YOLOv4 (You Only Look Once) deep learning model to detect defects and select the best method for glass lens defect detection. The final result is an industrial camera with high angle ring light and YOLOv4 as the main deep learning network to develop the system, and the accuracy of the glass lens defects reached 80.7%. It helps to improve the production process and enhance the performance of the automated production line.
摘要 i
ABSTRACT ii
誌謝 iii
目錄 iv
表目錄 viii
圖目錄 ix
第1章 緒論 1
1.1 研究背景 1
1.2 研究動機與目的 3
1.3 文獻探討 4
1.4 論文架構 6
第2章 基本理論介紹 7
2.1 相機成像 7
2.2 CMOS感光元件曝光方式 7
2.3 機器學習介紹 9
2.3.1 監督式學習 10
2.3.2 非監督式學習 10
2.3.3 半監督式學習 10
2.3.4 增強式學習 11
2.4 類神經網路(Neural Network) 11
2.4.1 神經元、銓重與偏差值(Neuron、Weight and Bias) 12
2.4.2 激勵函數(Activation Function) 13
2.4.3 損失函數(Loss Function) 15
2.4.4 優化器(Optimizer) 15
2.5 卷積神經網路(Convolutional Neural Network) 16
2.5.1 卷積層(Convolution Layer) 17
2.5.2 池化層(Pooling Layer) 17
2.5.3 平坦層(Flatten Layer) 18
2.5.4 全連接層(Fully Connected Layer) 19
第3章 硬體架構 20
3.1 相機 20
3.1.1 筆型顯微鏡 20
3.1.2 工業相機 21
3.2 鏡頭 22
3.2.1 1倍遠心鏡頭 23
3.2.2 0.3倍遠心鏡頭 24
3.3 光源 26
3.4 取像架構 27
3.4.1 取像架構第一版 27
3.4.2 取像架構第二版 28
3.5 代測物分析 29
3.5.1 玻璃透鏡 29
3.5.2 瑕疵分析 30
3.6 旋轉機構 32
3.6.1 Arduino UNO 34
3.6.2 驅動器DRV8825 35
3.6.3 治具 36
3.7 拍攝方式 37
3.8 拍攝結果 42
3.8.1 點光源 42
3.8.2 條光源 45
3.8.3 面光源 47
3.8.4 高角度環光 49
3.8.5 雙高角度環光 53
3.8.6 筆型顯微鏡與工業相機之比較 56
3.8.7 工業相機鏡頭之比較 57
第4章 軟體架構 60
4.1 YOLO 60
4.1.1 LabelImg 60
4.1.2 計算Anchor Boxes 61
4.1.3 CSPDarknet53 62
4.2 YOLOv4深度學習網路模型 63
4.3 YoloCSharpWrapper 64
4.4 YOLOv4資料集 65
4.5 瑕疵分類 66
第5章 實驗結果 67
5.1 YOLOv4 67
5.1.1 訓練結果 67
5.1.2 測試結果 68
5.2 系統開發 72
5.2.1 系統環境開發 72
5.2.2 流程設計 74
5.2.3 UI設計 74
第6章 結論與未來展望 76
6.1 結論 76
6.2 未來展望 76
參考文獻 77
[1]C. Phua and L. B. Theng, "Semiconductor wafer surface: Automatic defect classification with deep CNN," in 2020 IEEE REGION 10 CONFERENCE (TENCON), 2020: IEEE, pp. 714-719.
[2]X. Tao, D. Zhang, Z. Wang, X. Liu, H. Zhang, and D. Xu, "Detection of Power Line Insulator Defects Using Aerial Images Analyzed With Convolutional Neural Networks," IEEE Transactions on Systems, Man, and Cybernetics: Systems, vol. 50, no. 4, pp. 1486-1498, 2020.
[3]L. Shang, Q. Yang, J. Wang, S. Li, and W. Lei, "Detection of rail surface defects based on CNN image recognition and classification," in 2018 20th International Conference on Advanced Communication Technology (ICACT), 11-14 Feb. 2018 2018, pp. 45-51.
[4]M. Ferguson, R. Ak, Y. T. Lee, and K. H. Law, "Automatic localization of casting defects with convolutional neural networks," in 2017 IEEE International Conference on Big Data (Big Data), 11-14 Dec. 2017 2017, pp. 1726-1735.
[5]T. d. O. Simoes, J. C. Souza, J. D. S. d. Almeida, A. C. Silva, and A. C. d. Paiva, "Automatic Ocular Alignment Evaluation for Strabismus Detection Using U-NET and ResNet Networks," in 2019 8th Brazilian Conference on Intelligent Systems (BRACIS), 15-18 Oct. 2019 2019, pp. 239-244.
[6]Z. Li, J. Li, and W. Dai, "A Two-Stage Multiscale Residual Attention Network for Light Guide Plate Defect Detection," IEEE Access, vol. 9, pp. 2780-2792, 2021.
[7]X. Jiang, X. f. Wang, and D. Chen, "Research on Defect Detection of Castings Based on Deep Residual Network," in 2018 11th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI), 13-15 Oct. 2018 2018, pp. 1-6.
[8]J. Deng, J. Liu, C. Wu, T. Zhong, G. Gu, and B. W. K. Ling, "A Novel Framework for Classifying Leather Surface Defects Based on a Parameter Optimized Residual Network," IEEE Access, vol. 8, pp. 192109-192118, 2020.
[9]X. Chen, Z. An, L. Huang, S. He, X. Zhang, and S. Lin, "Surface Defect Detection of Electric Power Equipment in Substation Based on Improved YOLOV4 Algorithm," in 2020 10th International Conference on Power and Energy Systems (ICPES), 25-27 Dec. 2020 2020, pp. 256-261.
[10]H. Xie, Y. Li, X. Li, and L. He, "A Method for Surface Defect Detection of Printed Circuit Board Based on Improved YOLOv4," in 2021 IEEE 2nd International Conference on Big Data, Artificial Intelligence and Internet of Things Engineering (ICBAIE), 26-28 March 2021 2021, pp. 851-857.
[11]X. Hu, Z. Wei, and W. Zhou, "A video streaming vehicle detection algorithm based on YOLOv4," in 2021 IEEE 5th Advanced Information Technology, Electronic and Automation Control Conference (IAEAC), 12-14 March 2021 2021.
[12]S. Albahli, N. Nida, A. Irtaza, M. H. Yousaf, and M. T. Mahmood, "Melanoma Lesion Detection and Segmentation Using YOLOv4-DarkNet and Active Contour," IEEE Access, vol. 8, pp. 198403-198414, 2020.
[13]J. Y. Sung, S. B. Yu, and S. h. P. Korea, "Real-time Automatic License Plate Recognition System using YOLOv4," in 2020 IEEE International Conference on Consumer Electronics - Asia (ICCE-Asia), 1-3 Nov. 2020 2020, pp. 1-3.
[14]N. N. a. S. Natsupakpong, "Lens Quality Inspection using Image Processing and Machine Learning," in Management Science and Industrial Engineering, Osaka, Japan, 2020, pp. 184-188.
[15]L. W.-C. K. Chung-Feng Jeffrey, H. Yan-Ru , T. Hsin-Yang, L. Chi-Lung, W. Han-Cheng, "Automated defect inspection system for CMOS image sensor withmicro multi-layer non-spherical lens module," Journal of Manufacturing Systems, pp. 248-259, October 2017.
電子全文 電子全文(網際網路公開日期:20270803)
QRCODE
 
 
 
 
 
                                                                                                                                                                                                                                                                                                                                                                                                               
第一頁 上一頁 下一頁 最後一頁 top
無相關期刊