臺灣博碩士論文加值系統

隨著科技的不斷進步，自動化檢測技術已成為工業製造中不可或缺的一環。而在球墨鑄鐵製造過程中，石墨球化是一個關鍵步驟，能夠提高金屬材料的塑性和韌性。然而，金屬石墨球化率對產品的品質和性能具有極大的影響，因此需要對其進行精確檢測。目前，大多數金屬石墨球化率檢測仍依賴人工進行，這種方法不僅效率低下，還容易出現人為錯誤。為了提高檢測效率和精度，本研究將語義分割技術應用於金屬石墨球化率檢測中。語義分割技術可以自動標記影像中的不同物體和區域，並將它們劃分為不同的類別。在金屬石墨球化率檢測中，研究人員可以使用語義分割技術標記金屬材料中的石墨球和其他物質，從而計算石墨球的比例，進而確定金屬石墨球化率。實驗結果顯示，語義分割技術能夠精確標記金屬材料中的石墨球和其他物質，並計算出金屬石墨球化率。相較於傳統的人工檢測方法，語義分割技術能夠大幅提高檢測效率和精度，同時降低了人為錯誤的風險。

With the continuous advancement of technology, automated detection has become an essential part of industrial manufacturing. In the manufacturing process of ductile iron, spheroidization is a critical step that can improve the plasticity and toughness of metallic materials. However, the spheroidization rate of metallic materials has a great influence on the quality and performance of products, so it needs to be accurately detected. Currently, most spheroidization rate detection still relies on manual methods, which are not only inefficient but also prone to human errors. To improve detection efficiency and accuracy, this study applies semantic segmentation technology to the detection of spheroidization rate in metallic materials. Semantic segmentation technology can automatically label different objects and regions in images and classify them into different categories. In the detection of spheroidization rate in metallic materials, researchers can use semantic segmentation technology to label graphite balls and other substances in metallic materials, calculate the proportion of graphite balls, and determine the spheroidization rate of metallic materials. Experimental results show that semantic segmentation technology can accurately label graphite balls and other substances in metallic materials and calculate the spheroidization rate of metallic materials. Compared with traditional manual detection methods, semantic segmentation technology can significantly improve detection efficiency and accuracy, while reducing the risk of human errors.

目 錄
摘要 i
Abstract ii
誌 謝 iii
目 錄 iv
表 目 錄 vi
圖 目 錄 vii
第一章 緒論 1
1.1研究背景與動機 1
1.2研究目的 1
1.3論文架構 2
第二章 文獻探討 3
2.1球墨鑄鐵 3
2.2 球化率的計算方式及標準 4
2.2.1球化率與圓度係數 5
2.2.2依圓度係數分類 5
2.3影像分割 7
2.3.1傳統的分割方法 7
2.3.2 使用網路模型分割 9
第三章 研究方法 11
3.1 影像前處理 11
3.1.1 基於Stable Diffusion的資料擴充 14
3.2 語義分割網路架構 16
3.2.1 SegNet 16
3.2.1 U-Net 17
3.2.2 U2-Net 18
3.2.3 DeepLabv3 19
3.2.4 DeepLabv3+ 20
3.2.5 SegFormer 21
第四章 研究結果與分析 24
4.1 資料集 24
4.2 環境配置 24
4.3 模型訓練 25
4.4 效能評估指標 27
4.5 驗證結果比較 30
4.6 實際測試比較 31
4.7 GUI介面設計 34
第五章 結論與未來展望 36
5.1結論 36
5.2 未來展望 36
參考文獻 37

[1]J.W. Martin, in Materials for Engineering (Third Edition), 2006 Available:https://www.sciencedirect.com/science/article/abs/pii/B9781845691578500034
[2]Badrinarayanan, V., Handa, A., & Cipolla, R. (2015). Segnet: A deep convolutional encoder-decoder architecture for robust semantic pixel-wise labelling. arXiv preprint arXiv:1505.07293.
[3]Ronneberger, O., Fischer, P., and Brox, T., "U-net: Convolutional networks for biomedical image segmentation", in Proceedings of the International Conference on Medical image computing and computer-assisted intervention, 2015: Springer.
[4]Chen, L. C., et al., “DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs”, arXiv, Jun, 2, 2016. Available: https://doi.org/10.48550/arXiv.1606.00915.
[5]ISO, Microstructure of cast irons -- Part 1: Graphite classification by visual analysis, 2017, Available: https://www.sis.se/en/produkter/metallurgy/ferrous-metals/irons/iso-945-12017/
[6]Qin, X., et al., “U2-Net: Going deeper with nested U-structure for salient object detection”, Pattern Recognition, vol. 106, 2020.
[7]Zeng, H., Peng, S. and Li, D., “Deeplabv3+ semantic segmentation model based on feature cross attention mechanism”, Journal of Physics: Conference Series. IOP Publishing, 2020.
[8]Xie, E., et al, “SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers” , arXiv, May. 31, 2021. Available: https://doi.org/10.48550/arXiv.2105.15203.
[9]Liu, Y., et al.,“PaddleSeg: A High-Efficient Development Toolkit for Image Segmentation”, arXiv, Jan. 15, 2021 Available: https://doi.org/10.48550/arXiv.2101.06175
[10]R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer. Highresolution image synthesis with latent diffusion models. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2022.
[11]AUTOMATIC111, 2023, stable-diffusion-webui, https://github.com/AUTOMATIC1111/stable-diffusion-webui(2023)
[12]US patent 2485760, Keith Millis, "Cast Ferrous Alloy", issued 1949-10-25
[13]"Ductile Iron Data - Section 2". www.ductile.org. Archived from the original on 2001-01-29.
[14]ASTM A247-19 Standard Test Method for Evaluating the Microstructure of Graphite in Iron Castings. (2019, February 22). ASTM COMPASS. https://www.astm.org/a0247-19.html
[15]GB/T 9441-2021 Metallographic Test Method for Spheroidal Graphite Cast Iron. (2021, December 31). https://openstd.samr.gov.cn/bzgk/gb/newGbInfo?hcno=D24E4E418C01F543351FA0A1F82D3659
[16]提升石墨球化率量測、評估的效率. (n.d.). Kyence. https://www.keyence.com.tw/ss/products/microscope/vhx-casestudy/automobile/graphite-spheroidization-rate.jsp
[17]劉平. (2004, February 26). 圖像分割閾值選取技術綜述. Rocky的部落格. https://rocky69.pixnet.net/blog/post/200429091
[18]自适应阈值（adaptiveThreshold）分割原理及实现. (2022, December 13). 古月居. https://www.guyuehome.com/41457
[19]区域生长算法原理及实现. (2022, December 11). 古月居. https://www.guyuehome.com/41458
[20]基于区域的图像分割——区域分裂与合并. (2020, January 19). GuoYi的博客. https://reurl.cc/11KxzD
[21]Level Set （水平集）算法是什么？. (2016, March 26). 知乎. https://www.zhihu.com/question/22608763
[22][Day24]什麼是機器學習Machine Learning？. (2018, January 12). IT邦幫忙. https://ithelp.ithome.com.tw/articles/10196922
[23]Labelme. (2023, May 16). GITHUB. https://github.com/wkentaro/labelme