中醫診斷時以望、聞、問、切作為辨證的重要依據。在望診中，中醫認為透過舌象的鑑別，可以瞭解人體五臟六腑、氣血的狀態。
人工智慧的發展為醫療領域帶來了許多應用，例如分析各種資訊以輔助醫師診斷。中醫亦然，由人工智慧分析舌象圖片，就能為醫師提供初步判讀的結果加速診斷、處方。訓練模型時需要使用到大量資料來提升模型辨識舌象的準確性，然而醫學影像的取得往往需牽涉到病人的隱私層面以及相關法律規範，取得大量舌象圖片也非易事。傳統的圖像資料擴增方式，透過對影像進行旋轉、平移、翻轉、縮放、伸展或是調整影像亮度對比的手法，能夠將既有的圖像資料有效的增加，但對於資料的整體分佈並無有效的加廣，而生成式學習的崛起，讓人工智慧得以創造近似於原資料集的假影像。
本研究透過應用多個生成對抗網路（Generative Adversarial Networks, GAN）的架構方法，利用原有的舌象圖片資料，嘗試生成逼真的舌象圖片，並觀察各方法所生成的圖片品質以及生成式學習是否能有效為舌象分類模型帶來幫助。

In traditional Chinese medicine, observation, olfaction, inquiry, and palpation are important foundations for diagnosis. During the observation, Chinese medicine believes that through the differentiation of tongue appearance, one can understand the condition of the body's organs, meridians, and the state of Qi and blood.
The development of artificial intelligence has brought many applications to the field of medicine, such as analyzing various information to assist doctors in diagnosis. The same applies to traditional Chinese medicine, where artificial intelligence can analyze tongue images to provide doctors with preliminary interpretations, accelerating diagnosis and prescription. Training models require a large amount of data to improve the accuracy of tongue image recognition. However, acquiring medical images often involves patient privacy and relevant legal regulations, making it difficult to obtain a large number of tongue images. Traditional image data augmentation techniques, such as rotating, translating, flipping, scaling, stretching, or adjusting image brightness and contrast, can effectively increase existing image data. However, they do not effectively broaden the overall distribution of the data. The rise of generative learning allows artificial intelligence to create fake images that are similar to the original dataset.
This study applies multiple generative adversarial network architectures, to utilize existing tongue image data and attempt to generate realistic tongue images. The study also observes the quality of the generated images by each method and whether generative learning can effectively assist tongue image classification models.

摘要 i
ABSTRACT ii
目錄 iii
圖目錄 vi
表目錄 viii
第一章 緒論 1
1-1 研究背景 1
1-2 研究動機與目的 3
1-3 論文架構 4
第二章 相關研究與文獻探討 5
2-1 自動化舌診系統與舌象分類 5
2-1-1 自動化舌診系統 5
2-1-2 舌象分類 5
2-2 資料擴增 7
2-3 生成對抗網路 8
2-3-1 DCGAN 12
2-3-2 WGAN/WGAN-GP 13
2-3-3 SNGAN 14
2-3-4 PGGAN（ProGAN） 15
2-4 生成樣本品質的評估方式 19
2-4-1 啟動分數（Inception Score） 19
2-4-2 Fréchet啟動距離（Fréchet Inception Distance） 20
2-4-3 結構相似性指數（Structural Similarity Index） 22
2-4-4 峰值信噪比（Peak Signal-to-Noise Ratio） 22
第三章 研究方法 23
3-1 研究設備與環境 23
3-2 資料集 25
3-3 生成圖片 26
3-3-1 圖片預處理 26
3-3-2 生成訓練 27
3-4 模型效能比較 31
第四章 實驗設計與結果討論 33
4-1 DCGAN 33
4-2 PGGAN 40
4-3 模型性能比較 49
4-3-1 FID 49
4-3-2 SSIM 49
4-3-3 PSNR 50
第五章 結論與未來展望 51
參考文獻 53

[1] J. Langr and V. Bok, GANs in Action - Deep learning with Generative Adversarial Networks. Manning, 2019.
[2] I. Goodfellow, "NIPS 2016 Tutorial: Generative Adversarial Networks," p. arXiv:1701.00160doi: 10.48550/arXiv.1701.00160.
[3] T. Karras, T. Aila, S. Laine, and J. Lehtinen, "Progressive Growing of GANs for Improved Quality, Stability, and Variation," p. arXiv:1710.10196doi: 10.48550/arXiv.1710.10196.
[4] I. J. Goodfellow et al., "Generative Adversarial Networks," p. arXiv:1406.2661doi: 10.48550/arXiv.1406.2661.
[5] 羅倫謙 and 蔣依吾, 現代臨床望舌. 臺北市: 知音出版社, 2019.
[6] C. Li, D. Zhang, and S. Chen, "Research about Tongue Image of Traditional Chinese Medicine(TCM) Based on Artificial Intelligence Technology," in 2020 IEEE 5th Information Technology and Mechatronics Engineering Conference (ITOEC), 12-14 June 2020 2020, pp. 633-636, doi: 10.1109/ITOEC49072.2020.9141557.
[7] X. Wang et al., "Artificial intelligence in tongue diagnosis: Using deep convolutional neural network for recognizing unhealthy tongue with tooth-mark," Computational and Structural Biotechnology Journal, vol. 18, pp. 973-980, 2020/01/01/ 2020, doi: https://doi.org/10.1016/j.csbj.2020.04.002.
[8] H. Weng, L. Li, H. Lei, Z. Luo, C. Li, and S. Li, "A weakly supervised tooth-mark and crack detection method in tongue image," Concurrency and Computation: Practice and Experience, vol. 33, no. 16, p. e6262, 2021, doi: https://doi.org/10.1002/cpe.6262.
[9] 黃英聰, "基於深度學習之中醫舌象辨識及研究," 碩士, 資訊工程學系, 國立東華大學, 花蓮縣, 2021.
[10] 李宸綾, "訓練物體辨識模型進行舌象辨識及圖像尺寸對模型判別能力之影響," 碩士, 資訊工程學系, 國立東華大學, 花蓮縣, 2022.
[11] C. Shorten and T. M. Khoshgoftaar, "A survey on Image Data Augmentation for Deep Learning," Journal of Big Data, vol. 6, no. 1, p. 60, 2019/07/06 2019, doi: 10.1186/s40537-019-0197-0.
[12] L. c. Lo, M. C. c. Hou, Y. l. Chen, J. Y. Chiang, and C. c. Hsu, "Automatic Tongue Diagnosis System," in 2009 2nd International Conference on Biomedical Engineering and Informatics, 17-19 Oct. 2009 2009, pp. 1-5, doi: 10.1109/BMEI.2009.5304910.
[13] Y. C. Hsu, Y. C. Chen, L. c. Lo, and J. Y. Chiang, "Automatic tongue feature extraction," in 2010 International Computer Symposium (ICS2010), 16-18 Dec. 2010 2010, pp. 936-941, doi: 10.1109/COMPSYM.2010.5685377.
[14] M.-C. Hu et al., "Automated tongue diagnosis on the smartphone and its applications," Computer Methods and Programs in Biomedicine, vol. 174, pp. 51-64, 2019/06/01/ 2019, doi: https://doi.org/10.1016/j.cmpb.2017.12.029.
[15] D. A. van Dyk and X.-L. Meng, "The Art of Data Augmentation," Journal of Computational and Graphical Statistics, vol. 10, no. 1, pp. 1-50, 2001/03/01 2001, doi: 10.1198/10618600152418584.
[16] T. DeVries and G. W. Taylor, "Dataset Augmentation in Feature Space," p. arXiv:1702.05538doi: 10.48550/arXiv.1702.05538.
[17] A. Radford, L. Metz, and S. Chintala, "Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks," p. arXiv:1511.06434doi: 10.48550/arXiv.1511.06434.
[18] S. Ioffe and C. Szegedy, "Batch normalization: Accelerating deep network training by reducing internal covariate shift," in International conference on machine learning, 2015: pmlr, pp. 448-456.
[19] J. T. Springenberg, A. Dosovitskiy, T. Brox, and M. Riedmiller, "Striving for Simplicity: The All Convolutional Net," p. arXiv:1412.6806doi: 10.48550/arXiv.1412.6806.
[20] M. Arjovsky, S. Chintala, and L. Bottou, "Wasserstein GAN," p. arXiv:1701.07875doi: 10.48550/arXiv.1701.07875.
[21] I. Gulrajani, F. Ahmed, M. Arjovsky, V. Dumoulin, and A. C. Courville, "Improved training of wasserstein gans," Advances in neural information processing systems, vol. 30, 2017.
[22] H. Zhang, I. Goodfellow, D. Metaxas, and A. Odena, "Self-attention generative adversarial networks," in International conference on machine learning, 2019: PMLR, pp. 7354-7363.
[23] A. Brock, J. Donahue, and K. Simonyan, "Large Scale GAN Training for High Fidelity Natural Image Synthesis," p. arXiv:1809.11096doi: 10.48550/arXiv.1809.11096.
[24] T. Salimans, I. Goodfellow, W. Zaremba, V. Cheung, A. Radford, and X. Chen, "Improved techniques for training gans," Advances in neural information processing systems, vol. 29, 2016.
[25] E. Denton, S. Chintala, A. Szlam, and R. Fergus, "Deep Generative Image Models using a Laplacian Pyramid of Adversarial Networks," p. arXiv:1506.05751doi: 10.48550/arXiv.1506.05751.
[26] C. Szegedy, V. Vanhoucke, S. Ioffe, J. Shlens, and Z. Wojna, "Rethinking the Inception Architecture for Computer Vision," p. arXiv:1512.00567doi: 10.48550/arXiv.1512.00567.
[27] M. Heusel, H. Ramsauer, T. Unterthiner, B. Nessler, and S. Hochreiter, "GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium," p. arXiv:1706.08500doi: 10.48550/arXiv.1706.08500.
[28] 財團法人國家實驗研究院. "國家高速網路與計算中心." https://www.nchc.org.tw/ (accessed 2023).
[29] 國家高速網路與計算中心. "TWCC臺灣AI雲." https://www.twcc.ai/ (accessed 2023).