隨著人工智慧日漸發展，工業界期待能將其導入產線，實現生產自動化的目標。本研究目標為開發一套深度學習系統，使用異常檢測 (Anomaly detection) 方法，在不需要異常資料的情況下仍能檢測出異常。另加上目標區域 (Region of Interest, ROI) 定位的方法，去除複雜的背景，提升模型效能。並應用於半導體製程中，晶圓檢測製程的針印自動化檢測。
此深度學習系統分為兩個階段。第一階段為目標區域定位，使用模板匹配的方式搜尋影像中目標檢測區域並將其定位；第二階段為異常檢測模型，根據第一階段定位的資訊，針對目標區域進行異常檢測之訓練，使用重建式的異常檢測模型，讀取輸入影像後重建出正常影像，再與輸入影像進行比對，找出異常區域，訓練資料為產線批量資料，不需異常資料，在判斷時仍能偵測出異常。
本研究在異常檢測中，針對四種不同的鋁墊資料集：全白、矽穿孔、深晶粒邊界與淺晶粒邊界中獲得準確度100%、97.9%、100%、98%的成績。最後也透過實驗，驗證了單純透過影像分割無法分辨出異常，以及背景與針印對異常檢測模型造成負面影響，此兩點更強化了本研究結合影像分割與異常檢測之重要性。

With the rapid development of artificial intelligence, it is expected to be implemented to mass production, achieving the goal of manufacturing automation. This research proposed an anomaly detection method trained without anomaly data, and it can detect anomaly. Besides, this research also combined with region of interest (ROI) localization. Remove the complex background to increase the performance of this method. This research implements to wafer testing in semiconductor process. Automatically detect the defects during testing.
There are two main processes in this research. The first part is ROI localization. This process aims to locate the position of target objects with template matching method. The second part is anomaly detection. It is based on the positions of target objects from the first part. The purpose is to ignore the background, training with target objects only. With reconstructed-based anomaly detection method, the model reads the input image, reconstructs an anomaly-free image, and compares both input and reconstructed image to find the anomaly area. Besides, this model could train only with normal data in mass production. However, it’s able to detect the anomaly data during testing.
This research reaches 100%, 97.9%, 100%, 98% accuracy on white, via, dark grain boundary and light grain boundary aluminum pads. Finally, this research also shows the importance of combining image segmentation and anomaly detection by experiments.

摘要 I
Abstract II
目錄 III
圖目錄 VI
表目錄 IX
第一章 緒論 1
1.1 前言 1
1.2 研究動機 2
第二章 文獻回顧 6
2.1 影像處理 6
2.2 機器學習 8
2.3 深度學習 9
2.3.1 神經元 10
2.3.2 類神經網路 10
2.3.3 卷積神經網路 11
2.4 影像分割 12
2.5 異常檢測 15
2.5.1 重建式異常檢測 16
2.5.2 嵌入式異常檢測 21
第三章 研究方法 26
3.1 資料集 26
3.1.1 晶圓影像種類 26
3.1.2 異常之定義 30
3.1.3 訓練、驗證與測試集 31
3.2 研究方法流程 32
3.3 ROI定位方法 34
3.4 資料前處理 35
3.5 異常檢測模型 36
3.5.1 異常模擬器 37
3.5.2 自編碼器架構 38
3.5.3 損失函數 39
3.5.4 異常判斷方法 40
3.6 效能評估指標 42
3.6.1 IoU 42
3.6.2 混淆矩陣 43
3.6.3 準確度、過殺率、漏放率、精確率、召回率與F1分數 44
3.7 消融研究 (Ablation Study) 46
3.8 實驗環境與規格 49
第四章 研究結果與討論 50
4.1 ROI定位結果 50
4.2 異常檢測訓練結果 52
4.3 消融研究結果 62
4.3.1 影像分割無法辨識壓傷異常 62
4.3.2 背景與針印對異常檢測模型效能之影響 64
第五章 結論 71
5.1 研究貢獻 71
5.2 未來展望 72
參考文獻 74

[1] C. M. Bishop, “Pattern Recognition and Machine Learning”. Information Science and Statistics. 2006: Springer New York, NY.
[2] J. Shlens, “A tutorial on principal component analysis”. arXiv preprint arXiv:1404.1100, 2014.
[3] M. A. Hall, Correlation-based feature selection for machine learning. 1999, The University of Waikato.
[4] Y. Lecun, Y. Bengio, and G. Hinton, “Deep learning”. Nature, 2015. 521(7553): pp. 436-444.
[5] V. Nair and G. E. Hinton. “Rectified linear units improve restricted boltzmann machines”. in Proceedings of the 27th international conference on machine learning. 2010.
[6] Y. Lecun, L. Bottou, Y. Bengio, and P. Haffner, “Gradient-based learning applied to document recognition”. Proceedings of the IEEE, 1998. 86(11): pp. 2278-2324.
[7] H. Lee, R. Grosse, R. Ranganath, and A. Y. Ng. “Convolutional deep belief networks for scalable unsupervised learning of hierarchical representations”. in Proceedings of the 26th annual international conference on machine learning. 2009.
[8] O. Ronneberger, P. Fischer, and T. Brox. “U-net: Convolutional networks for biomedical image segmentation”. in International Conference on Medical image computing and computer-assisted intervention. 2015. Springer.
[9] K. He, G. Gkioxari, P. Dollár, and R. Girshick. “Mask r-cnn”. in Proceedings of the IEEE international conference on computer vision. 2017.
[10] S. Ren, K. He, R. Girshick, and J. Sun, “Faster r-cnn: Towards real-time object detection with region proposal networks”. Advances in neural information processing systems, 2015. 28: pp. 91-99.
[11] T.-Y. Lin, P. Dollár, R. Girshick, K. He, et al. “Feature pyramid networks for object detection”. in Proceedings of the IEEE conference on computer vision and pattern recognition. 2017.
[12] G. E. Hinton and R. R. Salakhutdinov, “Reducing the dimensionality of data with neural networks”. Science, 2006. 313(5786): pp. 504-507.
[13] D. P. Kingma and M. Welling, “Auto-encoding variational bayes”. arXiv preprint arXiv:1312.6114, 2013.
[14] J. An and S. Cho, “Variational autoencoder based anomaly detection using reconstruction probability”. Special Lecture on IE, 2015. 2(1): pp. 1-18.
[15] V. Zavrtanik, M. Kristan, and D. Skočaj, “Reconstruction by inpainting for visual anomaly detection”. Pattern Recognition, 2021. 112: 107706.
[16] P. Bergmann, M. Fauser, D. Sattlegger, and C. Steger. “MVTec AD--A comprehensive real-world dataset for unsupervised anomaly detection”. in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2019.
[17] V. Zavrtanik, M. Kristan, and D. Skočaj. “Draem-a discriminatively trained reconstruction embedding for surface anomaly detection”. in Proceedings of the IEEE/CVF International Conference on Computer Vision. 2021.
[18] M. Cimpoi, S. Maji, I. Kokkinos, S. Mohamed, et al. “Describing textures in the wild”. in Proceedings of the IEEE conference on computer vision and pattern recognition. 2014.
[19] K. Perlin, “An image synthesizer”. ACM Siggraph Computer Graphics, 1985. 19(3): pp. 287-296.
[20] J. Ho, A. Jain, and P. Abbeel, “Denoising diffusion probabilistic models”. Advances in neural information processing systems, 2020. 33: pp. 6840-6851.
[21] J. Wyatt, A. Leach, S. M. Schmon, and C. G. Willcocks. “Anoddpm: Anomaly detection with denoising diffusion probabilistic models using simplex noise”. in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2022.
[22] A. Mousakhan, T. Brox, and J. Tayyub, “Anomaly Detection with Conditioned Denoising Diffusion Models”. arXiv preprint arXiv:2305.15956, 2023.
[23] P. Bergmann, M. Fauser, D. Sattlegger, and C. Steger. “Uninformed students: Student-teacher anomaly detection with discriminative latent embeddings”. in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2020.
[24] S. Wang, L. Wu, L. Cui, and Y. Shen. “Glancing at the patch: Anomaly localization with global and local feature comparison”. in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2021.
[25] T. Defard, A. Setkov, A. Loesch, and R. Audigier. “Padim: a patch distribution modeling framework for anomaly detection and localization”. in International Conference on Pattern Recognition. 2021. Springer.
[26] K. Roth, L. Pemula, J. Zepeda, B. Schölkopf, et al. “Towards total recall in industrial anomaly detection”. in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2022.
[27] R. E. Devor, T. Chang, and J. W. Sutherland, “Statistical Quality Design and Control: Contemporary Concepts and Methods”. 1992: Macmillan.