目前主流的 AI 技術多半依賴高性能之電腦主機，而運行於微控制器上之嵌 入式系統並不多見。若是能將已趨成熟之 AI 技術運行於微控制器上，不僅可以 讓 AI 技術在更多領域應用，還可在許多應用中帶來可觀的功耗的節約。本研究 藉由加入了虛擬機器來使微控制器獲得靈活佈署的特性，從而使實作於微控制 器之深度神經網路應用可用更方便的方式佈署。在彌補了微控制器與電腦主機 應用深度神經網路在便利性上的差距後，前述之目標將變得切實可行。接著進 行的實驗中也驗證了虛擬機器的功能以及實作神經網路的正確性。本研究在將 深度神經網路應用帶進微控制的嵌入式系統中的同時，保留了微控制器的優 點，並克服了它的缺點。與高耗能高成本主機相比，本研究極低的功率消耗將 帶來可觀的節能效益。

At present, mainstream AI technology mostly relies on high-performance computer hosts, and embedded systems running on microcontrollers are rare. If the mature AI technology can be run on the microcontroller, you can not only let AI technology be applied in more fields, but also bring considerable power saving in many applications. In this research, the virtual machine was added to enable the microcontroller to be flexibly deployed, so that the deep neural network application implemented in the microcontroller can be deployed in a more convenient manner. After making up for the gap in the convenience of the deep neural network between the microcontroller and the host computer, the aforementioned goals will become feasible. The subsequent experiments also verified the function of the virtual machine and the correctness of the implemented neural network. This research preserves the advantages of the microcontroller and overcomes its shortcomings while bringing deep neural network applications into the microcontroller-based embedded system. Compared with high-energy and high-cost hosts, the extremely low power consumption system of this research will bring considerable energy savings.

參考文獻
[1] B. Chen, J. Wan, L. Shu, P. Li, M. Mukherjee, and B. J. I. A. Yin, "Smart factory of industry 4.0: key technologies, application case, and challenges," vol. 6, pp. 6505-6519, 2018.
[2] R. Jain, R. Kasturi, and B. G. Schunck, Machine vision. McGraw-Hill New York, 1995.
[3] E. R. Davies, Machine vision: theory, algorithms, practicalities. Elsevier, 2004.
[4] M. Sonka, V. Hlavac, and R. Boyle, Image processing, analysis, and machine vision. Cengage Learning, 2014.
[5] 謝欣蓓, "整合 GRAFCET 虛擬機器的智慧型控制器開發平台," 中央大學資訊工程學系學位論文, pp. 1-74, 2011.
[6] C.-W. Cheng, "整合可程式邏輯控制與機器視覺的 PLCopen 開發平台," National Central University, 2017.
[7] R. W. Lewis, Programming industrial control systems using IEC 1131-3 (no. 50). Iet, 1998.
[8] M. Maslar, "PLC standard programming languages: IEC 1131-3," in Pulp and Paper Industry Technical Conference, 1996., Conference Record of 1996 Annual, 1996: IEEE, pp. 26-31.
[9] A. Otto and K. J. I. I. E. M. Hellmann, "IEC 61131: A general overview and emerging trends," vol. 3, no. 4, 2009.
[10] M. Tiegelkamp and K.-H. John, IEC 61131-3: Programming industrial automation systems. Springer, 1995.
[11] E. V. d. Wal, "Creating reusable, hardware independent motion control applications via IEC 61131-3 and PLCopen Motion Control Profile," in 2001 IEEE/ASME International Conference on Advanced Intelligent Mechatronics. Proceedings (Cat. No.01TH8556), 8-12 July 2001 2001, vol. 2, pp. 769-774 vol.2, doi: 10.1109/AIM.2001.936763. [Online]. Available: https://ieeexplore.ieee.org/ielx5/7453/20266/00936763.pdf?tp=&arnumber=936763&isnumber=20266
[12] E. v. d. Wal, "Introduction into IEC 1131-3 and PLCopen," in IEE Colloquium on The Application of IEC 61131 to Industrial Control: Improve Your Bottom Line Through High Value Industrial Control Systems (Ref. No. 1999/076), 30-30 March 1999 1999, pp. 2/1-2/8, doi: 10.1049/ic:19990423. [Online]. Available: https://ieeexplore.ieee.org/ielx5/6267/16735/00771818.pdf?tp=&arnumber=771818&isnumber=16735
[13] T. Mitchell, B. Buchanan, G. DeJong, T. Dietterich, P. Rosenbloom, and A. Waibel, "Machine learning," Annual review of computer science, vol. 4, no. 1, pp. 417-433, 1990.
[14] E. P. Luke, D. W. Creavin, and R. R. Reetz, "System and method for automatic optical inspection," ed: Google Patents, 1996.
[15] R. Kauth, A. Pentland, and G. Thomas, "Blob: An unsupervised clustering approach to spatial preprocessing of mss imagery," 1977.
[16] J. M. Zurada, Introduction to artificial neural systems. West publishing company St. Paul, 1992.
[17] H. Bourlard and Y. Kamp, "Auto-association by multilayer perceptrons and singular value decomposition," Biological cybernetics, vol. 59, no. 4-5, pp. 291-294, 1988.
[18] R. Hecht-Nielsen, "Theory of the backpropagation neural network," in Neural networks for perception: Elsevier, 1992, pp. 65-93.
[19] T. Joachims, "Text categorization with support vector machines: Learning with many relevant features," in European conference on machine learning, 1998: Springer, pp. 137-142.
[20] D. P. Kingma and J. Ba, "Adam: A method for stochastic optimization," arXiv preprint arXiv:1412.6980, 2014.
[21] J. Chen and Q. Gu, "Padam: Closing the generalization gap of adaptive gradient methods in training deep neural networks," 2018.
[22] N. Srivastava, G. Hinton, A. Krizhevsky, I. Sutskever, and R. Salakhutdinov, "Dropout: a simple way to prevent neural networks from overfitting," The Journal of Machine Learning Research, vol. 15, no. 1, pp. 1929-1958, 2014.
[23] X. Glorot and Y. Bengio, "Understanding the difficulty of training deep feedforward neural networks," in Proceedings of the thirteenth international conference on artificial intelligence and statistics, 2010, pp. 249-256.
[24] K. He, X. Zhang, S. Ren, and J. Sun, "Delving deep into rectifiers: Surpassing human-level performance on imagenet classification," in Proceedings of the IEEE international conference on computer vision, 2015, pp. 1026-1034.
[25] S. Ioffe and C. Szegedy, "Batch normalization: Accelerating deep network training by reducing internal covariate shift," arXiv preprint arXiv:1502.03167, 2015.