臺灣博碩士論文加值系統

頻域光學同調斷層掃描(fourier domain optical coherence tomography, FDOCT)是一種高解析度的光學成像技術，透過光學干涉、光譜分析和掃描方
式，獲得生物和材料系統內部微觀結構的橫截面斷層影像。具有高解析度、
深度資訊、實時成像和非侵入性等優勢，透過此訊息可以應用於多個領域，
如:人體疾病病變、透光材質的瑕疵檢測等，解決無法得知橫切面等問題。
近年來，深度學習在各個領域迅速崛起，取得了引人注目的成就。本
研究以自編碼器(autoencoder)為基礎，探討如何實現目標資料的自主標籤
編碼。透過針對不同 FD-OCT 的深度資訊(depth profile)樣本，分別建立了
多層感知機(multilayer perceptron, MLP)、卷積神經網路(convolutional neural
network, CNN)和殘差神經網路(residual neural network, ResNet) 的模型，以
進行特徵提取。藉由自編碼器的特性使得對深度資訊樣本進行有效的壓縮
和還原成編碼形式。在實驗中，我們嘗試調整不同隱向量輸出，以取得還
原和壓縮比的平衡。實驗結果顯示，在不同樣本下，不同的自編碼器具有
各自的優勢。實驗結果證明了自編碼器能夠有效地重建和還原 FD-OCT 深
度資訊樣本，並提供了一種新的編碼系統。這為未來的研究提供了新的方
向和可能性。

Fourier domain optical coherence tomography (FD-OCT) is a high-resolution optical
imaging technique that uses optical interference, spectral analysis, and scanning to obtain crosssectional images of the internal microstructure of biological and material systems. It has high
resolution, depth information, real-time imaging, and non-invasiveness advantages. This
information can be applied to multiple fields, such as human disease diagnosis and defect
detection of transparent materials, to solve problems that cannot be obtained by cross-section.
Deep learning has rapidly risen in various fields in recent years and achieved remarkable
results. This study is based on autoencoders to explore attaining autonomous label encoding of
target data. Feature extraction was performed by constructing multilayer perceptron (MLP),
convolutional neural network (CNN), and residual neural network (ResNet) models for
different FD-OCT depth profile samples. The characteristics of the autoencoder make it
possible to effectively compress and restore the depth information samples into an encoded
form. We tried adjusting different latent vector outputs in the experiment to balance the
reconstruction and compression ratio. The experimental results show that different
autoencoders have their advantages under different samples. The experimental results
demonstrate that the autoencoder can effectively reconstruct and restore FD-OCT depth
information samples and provides a new coding system. This offers new directions and
possibilities for future research.

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 前言....1
1.2 動機與目的....1
1.3 論文架構....2
第二章 研究背景及相關知識....3
2.1 頻域光學同調斷層掃描....3
2.2 類神經網路介紹....4
2.2.1 神經元（Neuron）....4
2.2.2 權重（Weight）....5
2.2.3 激勵函數（Activation Function）....5
2.2.4 層次結構（Layered Structure）....6
2.2.5 前向傳播（Forward Propagation）....7
2.2.6 反向傳播（Backpropagation）....7
2.2.7 損失函數（Loss Function）....7
2.2.8 訓練（Training）....8
2.3 卷積神經網路....8
2.3.1 Convolutional 卷積....9
2.3.2 Padding 填充....9
2.3.3 Pooling 池化層....10
2.4 殘差卷積神經網路....11
2.5 AUTOENCODER....12
第三章 研究內容與模擬實驗架構....13
3.1 實驗架構....13
3.2 單層結構....14
3.2.1 傳統的FD-OCT深度資訊訓練資料....14
3.2.2 無偽影FD-OCT深度資訊訓練資料....14
3.2.3 Spectrum訓練資料....14
3.3 雙層結構....15
3.3.1 傳統的FD-OCT深度資訊訓練資料....15
3.3.2 無偽影FD-OCT深度資訊訓練資料....15
3.3.3 在OSA端所獲取之頻譜Spectrum....15
3.4 驗證成功率說明:....16
3.4.1 閥值示意圖:....16
3.4.2 距離判定示意圖:....17
3.4.3 還原成功示意圖:....17
3.4.4 還原失敗示意圖:....18
第四章 模型架構及數據的分析....19
4.1 神經網路訓練MLP模型架構圖....19
4.1.1 MLP模型一:....19
4.1.2 MLP模型二: ....21
4.1.3 MLP模型三: ....23
4.2 神經網路訓練CNN模型架構圖....25
4.2.1 CNN模型一:....25
4.2.2 CNN模型二:....27
4.2.3 CNN模型三:....29
4.3 神經網路訓練RESNET模型架構圖....31
4.3.1 ResNet模型一:....31
4.3.2 ResNet模型二:....33
4.3.3 ResNet模型三:....35
4.4 實驗結果比較....37
第五章 結論....40
5.1 論文結論: 40
5.2 未來研究: 40
參考文獻 41

[1]M. F. Shirazi, R. E. Wijesinghe, N. K. Ravichandran, P. Kim, M. Jeon, and J. Kim, "Optical thin film inspection using parallel spectral domain optical coherence tomography," in 2017 25th Optical Fiber Sensors Conference (OFS), 2017, pp. 1-4.
[2]C. Flueraru, Y. Mao, S. Sherif, and S. Chang, "Fourier Domain - Optical Coherence Tomography based on a Quadrature Mach-Zehnder Interferometer," in 2007 29th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, 2007, pp. 2643-2646.
[3]A. Shrestha and A. Mahmood, "Review of Deep Learning Algorithms and Architectures," IEEE Access, vol. 7, pp. 53040-53065, 2019.
[4]K. Alsaih, M. Z. Yusoff, T. B. Tang, I. Faye, and F. Mériaudeau, "Performance Evaluation Of Convolutions And Atrous Convolutions In Deep Networks For Retinal Disease Segmentation On Optical Coherence Tomography Volumes," in 2020 42nd Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC), 2020, pp. 1863-1866.
[5]X. Lv et al., "Defect Detection in Transparent Materials Based on Optical Coherence Tomography," in 2023 21st International Conference on Optical Communications and Networks (ICOCN), 2023, pp. 1-3.
[6]M. Krichen, "Generative Adversarial Networks," in 2023 14th International Conference on Computing Communication and Networking Technologies (ICCCNT), 2023, pp. 1-7.
[7]T. B. Brown et al., "Language Models are Few-Shot Learners," p. arXiv:2005.14165Accessed on: May 01, 2020. doi: 10.48550/arXiv.2005.14165 Available: https://ui.adsabs.harvard.edu/abs/2020arXiv200514165B
[8]P. Isola, J.-Y. Zhu, T. Zhou, and A. A. J. a. e.-p. Efros, "Image-to-Image Translation with Conditional Adversarial Networks," p. arXiv:1611.07004Accessed on: November 01, 2016. doi: 10.48550/arXiv.1611.07004 Available: https://ui.adsabs.harvard.edu/abs/2016arXiv161107004I
[9]J.-Y. Zhu, T. Park, P. Isola, and A. A. J. a. e.-p. Efros, "Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks," p. arXiv:1703.10593Accessed on: March 01, 2017. doi: 10.48550/arXiv.1703.10593 Available: https://ui.adsabs.harvard.edu/abs/2017arXiv170310593Z
[10]Y. Choi, M. Choi, M. Kim, J. W. Ha, S. Kim, and J. Choo, "StarGAN: Unified Generative Adversarial Networks for Multi-domain Image-to-Image Translation," in 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018, pp. 8789-8797.
[11]H.-C. Cheng, J.-F. Huang, and Y.-H. Hsieh, "Numerical analysis of one-shot full-range FD-OCT system based on orthogonally polarized light," Optics Communications, vol. 282, no. 14, pp. 3040-3045, 2009/07/15/ 2009.
[12]C. Yu, Y. H. Li, Y. H. Tseng, and Y. M. Chen, "Low-Latency LeNet-5 Architecture for Handwritten Digit Recognition," in 2023 IEEE International Conference on Consumer Electronics-Asia (ICCE-Asia), 2023, pp. 1-3.
[13]Y. Lecun, L. Bottou, Y. Bengio, and P. Haffner, "Gradient-based learning applied to document recognition," Proceedings of the IEEE, vol. 86, no. 11, pp. 2278-2324, 1998.
[14]K. He, X. Zhang, S. Ren, and J. Sun, "Deep Residual Learning for Image Recognition," in 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016, pp. 770-778.
[15]K. He, X. Zhang, S. Ren, and J. J. a. e.-p. Sun, "Deep Residual Learning for Image Recognition," p. arXiv:1512.03385Accessed on: December 01, 2015. doi: 10.48550/arXiv.1512.03385 Available: https://ui.adsabs.harvard.edu/abs/2015arXiv151203385H
[16]G. E. Hinton and R. R. Salakhutdinov, "Reducing the Dimensionality of Data with Neural Networks," vol. 313, no. 5786, pp. 504-507, 2006.
[17]C.-Y. Liou, W.-C. Cheng, J.-W. Liou, and D.-R. Liou, "Autoencoder for words," Neurocomputing, vol. 139, pp. 84-96, 2014/09/02/ 2014.
[18]N. Alamsyah, M. N. Fauzan, A. G. Putrada, and S. F. Pane, "Autoencoder Image Denoising to Increase Optical Character Recognition Performance in Text Conversion," in 2022 International Conference on Advanced Creative Networks and Intelligent Systems (ICACNIS), 2022, pp. 1-6.
[19]F. Barreto, S. Yadav, S. Patnaik, and J. Sarvaiya, "SIFT Features for Deep and Variational Autoencoders: A Performance Comparison," in 2020 2nd International Conference on Advances in Computing, Communication Control and Networking (ICACCCN), 2020, pp. 652-655.