近年來已有許多研究將深度學習應用於光學領域和影像處理，且皆有一定程度的進展。在光學領域中，像差是光學元件不可不考慮的關鍵因素之一，像差用來描述光的理想波前變形的狀況，人們從過去希望減少光學元件中像差的產生，到現在已經可以利用特定像差形式達到所需的目的，而這些都要從像差的量測開始。過去已有許多研究利用公式生成理想干涉條紋圖片作為神經網路模型的訓練資料，將深度學習應用於像差係數的預測，為的就是減少從干涉條紋圖片轉換至像差係數所需的複雜計算和時間。然而實際情況下的干涉條紋圖片帶有許多雜訊和缺陷，這些都會使得神經網路模型難以使用真實干涉條紋圖片進行像差係數的預測。
本篇研究架設麥克森干涉儀並搭配空間光調製器調製光波相位，收取帶有像差的干涉條紋圖片，利用先前使用理想干涉條紋圖片所訓練完成之三種深度學習網路模型，採用遷移式學習以減少重新訓練全新模型所需的大量時間，並分別比較CNN、GAN和RNN三種不同模型架構的遷移式學習結果，最終三種架構的模型皆能將訓練資料目標域，由理想資料轉為真實干涉條紋圖片，同時確立了真實干涉條紋圖片用於像差係數預測的可行性，並提高深度學習模型於不同干涉條紋圖片的泛用性。

In recent years, many studies have applied deep learning to the field of optics and image processing, and all have made some progress. Optical aberration is used to describe the ideal wavefront deformation of an optical system, and it is one of the key factors that must be considered in optical engineering. In the past, researchers tried to reduce the optical aberrations by using specific aberration formulas, and these all started with the measurement of optical aberrations. To reduce computational complexity and time consumption, many researchers used the ideal interference patterns generated by mathematical formulas to predict optical aberration coefficients using neural network models. However, in the actual situation, the interference fringes taken by an image sensor have many noises and defects, which will make it difficult for the neural network model to use the images to predict the coefficients accurately.
In this study, we set up a Michelson interferometer with a spatial light modulator which is used to generate real interference fringes with aberrations. The patterns were taken by an image sensor. We chose three different types of deep learning models, CNN, GAN and RNN, and used the ideal interference patterns as training data to determine the parameters. The parameters were used for transfer learning to save the time on retraining the models for using the real interference patterns. We finally converted the training data target domain from ideal patterns into real interference patterns and compared the performance of the three deep learning models (CNN, GAN and RNN). The results show that the performance of the three models are acceptable in predicting the aberration coefficients using real interference patterns.

[1] Y. Wang, H. Wang, Y. Li, C. Hu, H. Yang & M. Gu (2022). High‐accuracy, direct aberration determination using self‐attention‐armed deep convolutional neural networks. Journal of Microscopy, 286(1), 13-21.
[2] S. Zhang, L. Zhang, T. Gai, P. Gao, & Y. Wei (2021). Aberration Analysis and Control Based on Fully Connected Neural Network. 2021 China Semiconductor Technology International Conference (CSTIC).
[3] H. Jing, B. Fan, S. Wu, F. Wu, & T. Fan (2008). Measurement of optical surfaces with knife edge method. 3rd International Symposium on Advanced Optical Manufacturing and Testing Technologies: Optical Test and Measurement Technology and Equipment.
[4] J. Siv, R. Mayer, G. Beaugrand, G. Tison, R. Juvénal & G. Dovillaire (2019). Testing and characterization of challenging optics and optical systems with Shack Hartmann wavefront sensors. EPJ Web of Conference.
[5] S. De. Nicola, P. Ferraro, A. Finizio, & G.Pierattini (2001). Two-beam interferometer for measuring aberrations of optical components with axial symmetry. Applied Optics, 40(10), 1631-1636.
[6] K.H. Wan, X.L. Liao, M. Yu, Tsui, R. W., V.W. Chow, K.K. Chong & T.C. Chan (2022). Wavefront aberrometry repeatability and agreement—A comparison between Pentacam AXL Wave, iTrace and OPD‐Scan III. Ophthalmic and Physiological Optics.
[7] M. Servin, J.C. Estrada, & J.A. Quiroga, (2009). The general theory of phase shifting algorithms. Optics express, 17(24), 21867-21881.
[8] Y. Zhao, W. Yan, Y. Gao, Z. Yuan, Z.C. Ren, X.L. Wang, J. Ding & H.T. Wang (2021). High-Precision Calibration of Phase-Only Spatial Light Modulators. IEEE Photonics Journal, 14(1), 1-8.
[9] D. Yoo, S.W. Nam, Y. Jo, S. Moon, C.K. Lee & B. Lee (2022). Learning-based compensation of spatially varying aberrations for holographic display. JOSA A, 39(2), A86-A92.
[10] X. Hu, J. Lian, D. Zhang, X. Gao, L. Jiang & W. Chen (2022). Video anomaly detection based on 3D convolutional auto-encoder. Signal, Image and Video Processing, 1-9.
[11] X. Su, X. Yu, D. Chen, H. Li, X. Gao, X. Sang, X. Pei, X. Xie, Y. Wang & B. Yan (2022). Regional selection-based pre-correction of lens aberrations for light-field displays. Optics Communications, 505, 127510.
[12] X. Wu, J. Li, G. Zhou, B. Lü, Q. Li & H. Yang (2021). RRG-GAN Restoring Network for Simple Lens Imaging System. Sensors, 21(10), 3317.
[13] G. Côté, J.F. Lalonde & S. Thibault, (2019). Introducing a dynamic deep neural network to infer lens design starting points. Current Developments in Lens Design and Optical Engineering XX.
[14] S. Kim, W. Kim, Y.K. Noh & F.C. Park (2017). Transfer learning for automated optical inspection. 2017 international joint conference on neural networks (IJCNN).
[15] M.N. Vesperinas (2006). Scattering and diffraction in physical optics. World Scientific Publishing Company.
[16] B.H. Walker (1996). Forming Interference Fringes. Optics and Photonics News, 7(5), 34-37.
[17] G.M. Dai (2008). Wavefront optics for vision correction (Vol. 179). SPIE press.
[18] J. Perina (1985). Coherence of light. Springer Science & Business Media.
[19] B. Cao, C. Liu, L. Qin & X. Li (2021). Electroplating solution concentration detection based on interference principle and balanced detector. Optik, 240, 166903.
[20] P. Hariharan (1987). III Interferometry with Lasers. Progress in Optics, 24, 103-164.
[21] S. Wang, Q. Zhu & G. Cao (2009). Dispersed Rayleigh interferometer. 4th International Symposium on Advanced Optical Manufacturing and Testing Technologies: Optical Test and Measurement Technology and Equipment,
[22] M. Françon (1979). Optical interferometry. In Neutron interferometry.
[23] H. Hopkins (1955). Interferometric methods for the study of diffraction images. Optica Acta: International Journal of Optics, 2(1), 23-29.
[24] E.P. Goodwin & J.C. Wyant (2006). Field guide to interferometric optical testing.
[25] D. Malacara (1992). Twyman–green interferometer. Optical shop testing, 2, 46-96.
[26] K. Zetie, S. Adams & R. Tocknell (2000). How does a Mach-Zehnder interferometer work? Physics Education, 35(1), 46.
[27] D. Wang & R. Liang (2017). Point Diffraction Interferometry. In Optical Interferometry. IntechOpen.
[28] Y. Li, Y. Yang, C. Wang, Y. Chen & X. Chen (2017). Point diffraction in terference detection technology. Chin. Opt, 10, 391-414.
[29] M. Konnik & J. De Doná (2013). Waffle mode mitigation in adaptive optics systems: A constrained Receding Horizon Control approach. 2013 American Control Conference.
[30] V. Lakshminarayanan & A. Fleck (2011). Zernike polynomials: a guide. Journal of Modern Optics, 58(7), 545-561.
[31] T.A. Brummelaar, (1996). Modeling atmospheric wave aberrations and astronomical instrumentation using the polynomials of Zernike. Optics Communications, 132(3-4), 329-342.
[32] R. J. Noll (1976). Zernike polynomials and atmospheric turbulence. JOSA A, 66(3), 207-211.
[33] G. Lai & T. Yatagai (1991). Generalized phase-shifting interferometry. JOSA A, 8(5), 822-827.
[34] S. Zhang (2012). Composite phase-shifting algorithm for absolute phase measurement. Optics and Lasers in Engineering, 50(11), 1538-1541.
[35] G.M. Dai & V.N. Mahajan (2008). Orthonormal polynomials in wavefront analysis: error analysis. Applied Optics, 47(19), 3433-3445.
[36] L. Deng & D. Yu (2014). Deep learning: methods and applications. Foundations and trends® in signal processing, 7(3–4), 197-387.
[37] W. Rawat & Z. Wang (2017). Deep convolutional neural networks for image classification: A comprehensive review. Neural computation, 29(9), 2352-2449.
[38] S. Hochreiter & J. Schmidhuber (1997). Long short-term memory. Neural computation, 9(8), 1735-1780.
[39] I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville & Y. Bengio (2020). Generative adversarial networks. Communications of the ACM, 63(11), 139-144.
[40] J.J. Hopfield (1982). Neural networks and physical systems with emergent collective computational abilities. Proceedings of the national academy of sciences, 79(8), 2554-2558.
[41] Y. Yu, X. Si, C. Hu & J. Zhang (2019). A review of recurrent neural networks: LSTM cells and network architectures. Neural computation, 31(7), 1235-1270.
[42] X. Han, Z. Zhang, N. Ding, Y. Gu, X. Liu, Y. Huo, J. Qiu, Y. Yao, A. Zhang & L.Zhang (2021). Pre-trained models: Past, present and future. AI Open, 2, 225-250.
[43] J.W. Whang, Y.Y. Chen, C.M. Chang, Y.C. Liang, T.H. Yang, C.T. Lin & C.H. Chou (2020). Prediction technique of aberration coefficients of interference fringes and phase diagrams based on convolutional neural network. Optics express, 28(25), 37601-37611.
[44] P. Ballester & R.M. Araujo (2016). On the performance of GoogLeNet and AlexNet applied to sketches. Thirtieth AAAI conference on artificial intelligence.
[45] C. Szegedy, V. Vanhoucke, S. Ioffe, J. Shlens & Z. Wojna (2016). Rethinking the inception architecture for computer vision. Proceedings of the IEEE conference on computer vision and pattern recognition.
[46] J.W. Whang, Y.Y. Chen, T.H. Yang, C.T. Lin, Z.J. Jian & C.H. Chou (2021). Zernike coefficient prediction technique for interference based on generation adversarial network. Applied Sciences, 11(15), 6933.
[47] X. Shi, Z. Chen, H. Wang, D.Y. Yeung, W.K. Wong & W.C. Woo (2015). Convolutional LSTM network: A machine learning approach for precipitation nowcasting. Advances in neural information processing systems, 28.
[48] F. Chollet "Xception: Deep Learning with Depthwise Separable Convolutions," Computer Vision and Pattern Recognition, Oct 2016 , doi: 10.48550/arXiv.1610.02357