臺灣博碩士論文加值系統

本研究針對台灣橡膠產製造業中保護線性滑軌的橡膠墊片進行瑕疵檢測。由於傳統的檢測方式需要大量人力進行生產與品檢，因此本研究提出了一種基於卷積神經網路 (CNN) 的深度學習模型。本研究採用了三種預先訓練好的模型 (VGG16、InceptionV3 和 DenseNet)，瑕疵類型包括毛邊、未分離、破損、灌點破裂、油污、螺絲孔不良、疊料凸點不良、定位槽破裂及刮傷九種瑕疵，以評估模型的性能。
實驗結果表明，所提出的模型在檢測橡膠墊片瑕疵方面取得了較好的效果，其中 DenseNet 的準確率最高，達到了98.75％。此外，本研究還對模型的召回率和精確率進行了評估，三種模型在召回率和精確率方面均取得了較好的表現，其中 DenseNet 的召回率和精確率均達到了98.8％。有望應用於實際的生產場景中，降低人力成本、提高檢測效率，並提高產品的品質。

This study focuses on the detection of defects in rubber gaskets that protect linear slide rails in the Taiwanese rubber manufacturing industry. Traditional inspection methods require a large amount of labor for production and quality control, so this study proposes a deep learning model based on convolutional neural networks (CNNs). Three pre-trained models (VGG16, InceptionV3, and DenseNet) were used to evaluate the model's performance on nine types of defects, including rough edges, unshaved areas, damage, injection point rupture, oil stains, screw holes defects, overlapping material protrusions, positioning slot rupture, and scratches.
Experimental results show that the proposed model achieved good performance in detecting defects in rubber gaskets, with the highest accuracy achieved by DenseNet at 98.75%. In addition, the recall and precision of the models were evaluated, and all three models showed good performance in both recall and precision, with DenseNet achieving a recall and precision rate of 98.8%. The model is expected to be applied in actual production scenes to reduce labor costs, improve detection efficiency, and enhance product quality.

摘要...........................................................................i
Abstract......................................................................ii
誌謝.........................................................................iii
目錄..........................................................................iv
表目錄.........................................................................v
圖目錄........................................................................vi
第一章 緒論.....................................................................1
1-1 研究目的...................................................................1
1-2 研究背景...................................................................1
1-3 研究方法...................................................................3
1-4 論文大綱...................................................................4
第二章 橡膠墊片瑕疵資料收集及分類架構.............................................5
2-1 橡膠墊片瑕疵資料庫建立.......................................................5
2-2 橡膠墊片瑕疵種類............................................................8
2-3 橡膠墊片瑕疵樣本數據集......................................................12
2-4 橡膠墊片瑕疵影像優化.......................................................13
2-5 橡膠墊片瑕疵分類架構.......................................................14
第三章 橡膠墊片之Inception V3模型分析...........................................17
3-1 基於Inception V3模型對橡膠墊片瑕疵分類初步訓練及驗證.........................17
3-2 基於InceptionV3 模型對橡膠墊片初步測試結果..................................19
3-3 橡膠墊片瑕疵初步結果分析....................................................21
3-4 基於Inception V3模型對橡膠墊片瑕疵分類優化訓練及驗證.........................23
3-5 基於InceptionV3模型對橡膠墊片分類優化結果...................................24
第四章 橡膠墊片之DenseNet模型分析...............................................26
4-1 基於DenseNet模型對橡膠墊片瑕疵分類訓練及驗證.................................26
4-2 基於DenseNet 模型對橡膠墊片測試結果.........................................28
4-3 橡膠墊片瑕疵分類模型訓練與測試..............................................30
4-4 橡膠墊片深度學習模型分析與比較..............................................32
第五章 結論...................................................................33
參考文獻......................................................................34
Extended Abstract............................................................35

[1]顏俊杰，2022，”基於深度學習技術之玻璃瑕疵檢測”，國立高雄科技大學電機工程系碩士班碩士論文。
[2]黃俊欽，2022，”利用深度學習於車用 LED瑕疵檢測”，國立高雄科技大學電子工程系碩士班碩士論文。
[3]黃景暘，2021，”應用遷移學習於孿生網路之PCB瑕疵檢測”，國立陽明交通大學電子研究所碩士論文。
[4]吳漢威，2022，”基於機器學習之自動化檢測及應用於積層製造物件研究”，國立台灣科技大學自動化及控制研究所碩士論文。
[5]Mahardi,. 2021, “Development of Dogs and Cats Breed Classifier in Campus Area”, National Formosa University Department of Automation Engineering Master Thesis.
[6]Kaur, T. and Gandhi, T.K. “2019,Automated Brain Image Classification Based on VGG-16 and Transfer Learning”, 2019 International Conference on Information Technology (ICIT), pp. 94-98, Bhubaneswar, India,
[7]Szegedy, C., Vanhoucke, V., Ioffe, S., Shlens, J., and Wojna, Z., 2016, “Rethinking the Inception Architecture for Computer Vision”, In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (pp. 2818-2826).
[8]林國維，2022，”使用深度卷積神經網路應用於瑕疵影像分類”，國立臺北科技大學工業工程與管理系碩士班碩士論文。
[9]賴安祺，2022，”深度學習與遷移學習應用於即時花卉辨識系統之研究”，健行科技大學資訊工程系碩士班碩士論文。
[10]Huang, G., Liu, Z., Van Der Maaten, L., and Weinberger, K. Q., 2017, “Densely Connected Convolutional Networks”. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (pp. 4700-4708).
[11]蔡欣庭，2019，”利用DenseNet進行乳房影像分類之研究”，義守大學資訊工程學系碩士班碩士論文。
[12]Zhang, J.W., Lee, K.C. and Reddy, G.A.K. 2022 October 28-30, “Rubber Gasket Defect Classification by VGG16 model”, in 2022 IEEE Eurasia Conference on IOT, Communication and Engineering(ECICE), Yunlin, Taiwan.