隨著人工智慧(Artificial Intelligence, AI)的快速發展，將AI整合到其它科學領域已經成為推動技術創新的關鍵趨勢。在奈米光學領域，特別是奈米天線(nano-antenna)設計的研究中，AI技術已被用於提升奈米天線的性能與設計效率，其在增強無線通訊、光學感測、生物標記等皆具有廣泛的應用。
本篇論文利用有限差分時域(Finite-Difference Time-Domain, FDTD)方法模擬軟體 MEEP，進行電磁場計算與分析。我們設計出「I型」與矩形天線結構，將x極化的偶極子源(dipole source)放置於「I型」天線中心，藉由同時改變矩形天線結構與兩天線間距離，觀察其輻射模式的變化，除此之外，我們還建立了一套深度學習模型，分別為反向神經網路(Inverse Neural Network, INN)與前向神經網路(Forward Neural Network, FNN)以進行雙向預測。INN模型透過卷積與最大池化層進行特徵降採樣，擷取出天線輻射圖像的主要特徵，並通過全連接層進行最終的標籤分類，用以預測奈米天線的結構參數；FNN模型則以天線結構參數(分別為長度、寬度與間距)作為輸入，通過轉置卷積層將結構參數先進行上採樣的操作，進而在空間維度上擴展特徵圖，並重建出複雜的圖像細節，以預測出相對應的輻射模式。值得注意的是，在INN模型架構中，我們還應用了轉移學習中預訓練好的VGG16模型，允許在沒有大量數據集的情況下，有效地提升模型的學習與預測能力，能夠有效處理複雜的一對多映射問題，捕捉到更細微的數據特徵，進而提高模型整體的性能。
綜合上述，通過深度學習模型的迭代訓練與驗證，本研究期望在未來能夠取代傳統的光學模擬方法，避免繁瑣的實驗計算與長時間的模擬，不僅能夠高效準確地設計複雜的奈米天線結構，還能進一步提高模型預測的精確性、可靠性與泛化能力。

This thesis focuses on employing Artificial Intelligence (AI) in nanophotonics, specifically for nano-antenna design. AI has been utilized to enhance the design efficiency of nano-antennas and has found extensive applications in areas such as enhanced wireless communication, optical sensing, and biological markers.
Utilizing the simulation software-MEEP and the Finite-Difference Time-Domain (FDTD) method, this research conducts electromagnetic field analysis to design ‘I-shaped’ and rectangular antennas. By altering the structures and the distance between these antennas, changes in the far-field radiation pattern were analyzed. An inverse neural network (INN) and a forward neural network (FNN) are established for bi-directional prediction. The INN model uses convolution layers and max pooling layers for feature extraction from radiation patterns and employs a fully connected layer for the classification prediction of structural parameters. Conversely, the FNN model takes structural parameters as input and utilizes transposed convolution layers to effectively reconstruct far-field radiation patterns. It is noteworthy that in the INN model, we have also applied a pre-trained VGG16 model from transfer learning to enhance the accuracy of the model's predictions. Finally, we use a variety of error assessment metrics to conduct a comprehensive analysis of these models.

[1] D. Dregely, K. Lindfors, M. Lippitz, N. Engheta, M. Totzeck, H. Giessen, “Imaging and steering an optical wireless nanoantenna link,” Nature Communications, July (2014).

[2] 陳秉睿、陳國平, “奈米天線應用於薄膜太陽能電池吸收增強效益,”(2016).

[3] B. Wu, Y. Lou, D. Wu, Q. Min, X. Wan, H. Zhang, Y. Yu, J. Ma, G. Si, and Y. Pang,“Directivity-Enhanced Detection of a Single Nanoparticle Using a Plasmonic Slot Antenna,” Nano Letters, vol.22, pp.2374-2380, (2022).

[4]https://jasonblog.github.io/note/deep_learning/ji_qi_xue_7fd228_machine_learning__jie_shao.html

[5] D.E. Rumelhart, G.E. Hintont, and R.J. Williams, “Learning representations by backpropagating errors,” Nature, vol.323, pp.533-536, (1986).

[6] A. Krizhevsky, I. Sutskever, and G.E. Hinton, “ImageNet Classification with Deep Convolutional Neural Networks,” Communications of the ACM, vol.60, no.6, pp.84-90, (2017).

[7] M. Dampfhoffer ,T. Mesquida , A. Valentian , and L. Anghel, “Backpropagation-Based Learning Techniques for Deep Spiking Neural Networks: A Survey,” IEEE, pp.1-16, (2023).

[8] https://blogs.nvidia.com/blog/2016/07/29/whats-difference-artificial-intelligencemachine-learning-deep-learning-ai/

[9] R. P. Yuda, C. Aroef, Z. Rustam, and H. Alatas, “Gender Classification Based on Face Recognition using Convolutional Neural Networks (CNNs),” Journal of Physics: Conference Series, vol.1490, (2020).

[10] C. Dwork, F. McSherry, K. Nissim, and A. Smith, “Calibrating Noise to Sensitivity in Private Data Analysis,” In Theory of Cryptography Conference (TCC), vol.3876, pp.265-284, (2006).

[11] K.Wu, T. Rindzevicius, M.S. Schmidt, A.H. Thilsted, A. Boisen, “Optimizing silvercapped silicon nanopillars to simultaneously realize macroscopic, practical-level SERS signal reproducibility and high enhancement at low costs, Journal of Raman Spectroscopy, ” vol.48, pp.1808-1818, (2017).

[12] K.A. Willets and R.P. Van Duyne, “Localized Surface Plasmon Resonance Spectroscopy and Sensing,” Annual Review of Physical Chemistry, pp.267-270, (2007).

[13] 吳民耀、劉威志,“表面電漿子理論和模擬,” 物理雙月刊 28 卷 2 期 , pp.486-496, (2006).

[14] 邱國斌、蔡定平,“金屬表面電漿子簡介,” 物理雙月刊28卷2期, pp.473-485, (2006).

[15] N.W. Ashcroft and N.D. Mermin, “Solid state physics,”Academic Press Books, (2001).

[16] N.-G. Park, “Organometal Perovskite Light Absorbers Toward a 20% Efficiency LowCost Solid-State Mesoscopic Solar Cell,” The Journal of Physical Chemistry Letters, vol.4, pp.2423-2429, (2013).

[17] C.A. Balanis, “Antenna Theory: Analysis and Design,” John Wiley and Sons, (2016).

[18] K.M. See, “Photoluminescence Driven Gold Optical Yagi-Uda Nanoantenna as Directional Broadband Photon Source,” pp.48-52, (2016).

[19] M. Morshed, Z.Y. Li, B.C. Olbricht, L. Fu, A. Haque, L. Li, A.A. Rifat, M. Rahmani, A.E. Miroshnichenko & H.T. Hattori, “High Fluence Chromium and Tungsten Bowtie Nano-antennas,” Scientific Reports, (2019).

[20] K. S. Yee, “Numerical solution of initial boundary value problems involving maxwell's equations in isotropic media,” IEEE Transactions on Antennas and Propagation, vol. 14, pp.302-307, (1966).

[21] 藍永強、邱行偉, “電漿模擬,” 物理雙月刊廿八卷二期,pp.497-504, (2006).

[22] J. Berenger, “Perfectly matched layer for the FDTD solution of wavestructure interaction problemes,” IEEE Transactions on Antennas and Propagation, vol. 44, pp.110-117, (1996).

[23] E. Stevens, L. Antiga, T. Viehman 著, 黃駿譯, “核心開發者親授！PyTorch 深度學習攻略 (Deep Learning with Pytorch),” 旗標科技, CH.8, pp.2-57, (2021).

[24] 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).

[25] R. Afriandika, D.I. Mulyana, M. Ulum, and D. Yuliana, “Optimasi Image Classification Pada Hewan Ular Dengan Convolutional Neural Network,” Smart Comp, vol. 11, no. 2, pp. 270-282, (2022).

[26] “Introduction to Padding, https://www.geeksforgeeks.org/cnn-introduction-to-padding/

[27] L. Calderon J, “A new activation function for Neural Networks - logmoid”, Linkedin, (2021),

[28] S.R. Dubey, S.K. Singh, and B.B. Chaudhuri, “Activation functions in deep learning: A comprehensive survey and benchmark,” Elsevier, vol. 503, pp.92-108, (2022)

[29] “Activation Functions”, ML Glossary.https://ml-cheatsheet.readthedocs.io/en/latest/activation_functions.html?highlight=relu

[30] I. Goodfellow, Y. Bengio and A. Courville, “Deep Learning,” MIT, pp.67-69, (2016).

[31] S. V. G. Reddy, K.T. Reddy, V.V. Kumari, “Optimization of Deep Learning using Various Optimizers, Loss Functions and Dropout,” International Journal of Recent Technology and Engineering (IJRTE), vol.7, pp.448-455, (2019).

[32] Machine-Learning-Image-Classification, Jupyter Notebook.

[33] H. Noh, S. Hong, B. Han, “Learning Deconvolution Network for Semantic Segmentation,” IEEE International Conference on Computer Vision (ICCV), pp.1520-1528, (2015).

[34] F. Chollet, “Building powerful image classification models using very little data,”Keras, (2016).

[35] J. Wang, L. Bi, P. Sun, X. Jiao, X. Ma, X. Lei and Y. Luo, “Deep-Learning-Based Automatic Detection of Photovoltaic Cell Defects in Electroluminescence Images,”Sensors, vol.23, pp.297-318, (2022).

[36] A. Mumuni and F. Mumuni, “CNN Architectures for Geometric Transformation‑Invariant Feature Representation in Computer Vision: A Review,” SN Computer Science, pp.2-23, (2021)

[37] H. Taghvaee, A. Jain, X. Timoneda, C. Liaskos, S. Abadal, E. Alarcón and C.A. Albert,“Radiation Pattern Prediction for Metasurfaces: A Neural Network-Based Approach,”Sensors, vol.21, pp.2765-2790, (2021)