臺灣博碩士論文加值系統

近年來，研究人員對於更快捷方便的人機交互手段的探索與研究持續升溫。這一領域的研究主要集中在利用人體的神經訊號，如肌電訊號和腦波等，來實現對機器的控制，從而解放雙手，實現更便利的控制方式，在人機交互研究中，關鍵問題之一是訊號的處理和分析。研究人員使用深度學習和模式識別等技術，對肌電訊號和腦波訊號進行特徵提取和分類。這些技術能夠幫助識別出特定的運動模式和意識狀態，從而實現準確的機器控制。
本論文旨在利用肌電訊號作為人機交互的手段，並探索各種訊號處理方法，在結合了捲積神經網路後，在此領域的應用效果和潛力。我們評估了多種方法的表現，並找到了最優良的方法。此外，我們還引入了邊緣運算的概念，將最需要運算的辨識部份，移到具有GPU的電腦上來進行辨識，以提高辨識過程的速度和效能。

In recent years, there has been increasing interest and research in exploring more efficient and convenient human-machine interaction methods. This research field primarily focuses on utilizing human neural signals, such as electromyography (EMG) and electroencephalography (EEG), to enable control of machines, thus freeing the hands and achieving more convenient control methods. Signal processing and analysis are key challenges in human-machine interaction research. Researchers employ deep learning and pattern recognition techniques to extract features and classify EMG and EEG signals. These techniques facilitate the identification of specific movement patterns and conscious states, enabling accurate machine control.
The aim of this study is to leverage EMG signals as a means of human-machine interaction and explore various signal processing methods in conjunction with convolutional neural networks. We evaluated the performance of multiple methods and identified the most effective approach. Additionally, we introduced the concept of edge computing, where the computationally intensive recognition tasks are offloaded to computers with GPUs, thereby enhancing the speed and efficiency of the recognition process.

摘要 ............................................................................................................................................. i
Abstract ....................................................................................................................................... ii
誌謝 ........................................................................................................................................... iii
目錄 ........................................................................................................................................... iv
表目錄 ....................................................................................................................................... vi
圖目錄 ...................................................................................................................................... vii
1 第一章 緒論 .......................................................................................................................... 1
1.1 研究背景 .......................................................................................................................... 1
1.2 研究動機與目的 .............................................................................................................. 2
1.3 論文架構 .......................................................................................................................... 2
2 第二章 文獻探討 .................................................................................................................. 3
2.1 肌電訊號 .......................................................................................................................... 3
2.1.1 採集方法 .............................................................................................................. 3
2.1.2 肌電訊號種類 ...................................................................................................... 4
2.2 快速傅立葉轉換 .............................................................................................................. 5
2.2.1 短時傅立葉轉換 .................................................................................................. 6
2.3 深度學習 .......................................................................................................................... 7
2.3.1 CNN(Convolutional Neural Network) .................................................................. 8
2.3.2 前向傳播算法 ...................................................................................................... 9
2.3.3 反向傳播算法 .................................................................................................... 10
2.4 邊緣運算 ........................................................................................................................ 11
2.4.1 雲端運算 ............................................................................................................ 11
2.4.2 邊緣運算 ............................................................................................................ 12
2.4.3 Socket .................................................................................................................. 13
3 第三章 實驗方法與系統架構 ............................................................................................ 14
3.1 軟硬體介紹 .................................................................................................................... 14
3.1.1 Grove - EMG Detector ........................................................................................ 14
3.1.2 Raspberry pi 4B ................................................................................................... 15
3.1.3 邊緣運算電腦 .................................................................................................... 16
3.1.4 Duckiebot DB18 .................................................................................................. 16
3.2 手勢辨識 ........................................................................................................................ 17
v
3.2.1 數據收集 ............................................................................................................ 18
3.2.2 模型訓練流程 .................................................................................................... 22
3.2.3 模型架構 ............................................................................................................ 23
3.3 實驗系統架構 ................................................................................................................ 24
3.4 評估方法 ........................................................................................................................ 26
3.4.1 混淆矩陣 ............................................................................................................ 26
3.4.2 基於混淆矩陣的常用指標 ................................................................................ 27
3.4.3 Accuracy 與 loss 曲線 ......................................................................................... 28
4 第四章 實驗結果與討論 .................................................................................................... 29
4.1 肌電訊號辨識實驗結果 ................................................................................................ 29
4.1.1 方法 1 測試結果 ................................................................................................ 29
4.1.2 方法 2 測試結果 ................................................................................................ 30
4.1.3 方法 3 測試結果 ................................................................................................ 31
4.1.4 方法 4 測試結果 ................................................................................................ 32
4.1.5 總結 .................................................................................................................... 34
4.2 實際應用 ........................................................................................................................ 35
4.3 泛化能力測試 ................................................................................................................ 39
5 第五章 結論與未來研究 .................................................................................................... 41
參考文獻 .................................................................................................................................. 42
Extended Abstract ..................................................................................................................... 44

[1]Hoang, N. N., Lee, G. S., Kim, S. H., & Yang, H. J. (2019, September). Continuous hand gesture spotting and classification using 3D finger joints information. In 2019 IEEE International Conference on Image Processing (ICIP) (pp. 539-543). IEEE.
[2]Abdulghani, M. M., Franza, O., Fargo, F., & Raad, H. (2022, June). Brain Waves Pattern Recognition Using LSTM-RNN for Internet of Brain-Controlled Things (IoBCT) Applications. In 2022 IEEE International IOT, Electronics and Mechatronics Conference (IEMTRONICS) (pp. 1-5). IEEE.
[3]Shafivulla, M., Rajesh, V., & Khan, H. (2012, December). sEMG based human computer interface for robotic wheel. In 2012 4th International Conference on Intelligent Human Computer Interaction (IHCI) (pp. 1-5). IEEE.
[4]INFO研究圈. (2020). 一文詳解sEMG表面肌電 - 赢富儀器-上海心理人因|生物力學|駕駛行為研究設備一站式解决方案提供者. 取自：https://www.infoinstruments.cn/semg/(存取日期：2023年7月11日)
[5]Rawat, D., Pawar, L., Bathla, G., & Kant, R. (2022, August). Optimized Deep Learning Model for Lung Cancer Prediction Using ANN Algorithm. In 2022 3rd International Conference on Electronics and Sustainable Communication Systems (ICESC) (pp. 889-894). IEEE..
[6]El Haj, Y., El-Hag, A. H., & Ghunem, R. A. (2021). Application of deep-learning via transfer learning to evaluate silicone rubber material surface erosion. IEEE Transactions on Dielectrics and Electrical Insulation, 28(4), 1465-1467.
[7]葉明杰. (2004). 手臂運動之肌電訊號特性分析與應用 [碩士論文，國立交通大學，電機與控制工程系所]
[8]肌電電極2—一文了解高端肌電產品代表公司及其產品！_肌肉_區域_運動。 (2023). 取自：https://www.sohu.com/a/660047441_139908 (存取日期：2023年7月11日)
[9]Monnier, N., Ghali, D., & Liu, S. X. (2021, August). FFT and machine learning application on major chord recognition. In 2021 Twelfth International Conference on Ubiquitous and Future Networks (ICUFN) (pp. 426-429). IEEE.
[10]Ozdemir, M. A., Kisa, D. H., Guren, O., Onan, A., & Akan, A. (2020, November). EMG based hand gesture recognition using deep learning. In 2020 Medical Technologies Congress (TIPTEKNO) (pp. 1-4). IEEE.
[11]Allibhai, E. (2018). Building A Deep Learning Model using Keras. Towards Data Science. 取自：https://towardsdatascience.com/building-a-deep-learning-model-using-keras-1548ca149d37(存取日期：2023年7月11日)
 
[12]Wang, Q., & Wang, X. (2020, October). EMG-based Hand Gesture Recognition by Deep Time-frequency Learning for Assisted Living & Rehabilitation. In 2020 11th IEEE Annual Ubiquitous Computing, Electronics & Mobile Communication Conference (UEMCON) (pp. 0558-0561). IEEE.
[13]Basics of CNN in Deep Learning - Analytics Vidhya. (2022). 取自：https://www.analyticsvidhya.com/blog/2022/03/basics-of-cnn-in-deep-learning/(存取日期：2023年7月11日)
[14]雲端運算 - 維基百科，自由的百科全書。 (2023). 取自：https://zh.wikipedia.org/zh-tw/%E9%9B%B2%E7%AB%AF%E9%81%8B%E7%AE%97(存取日期：2023年7月11日)
[15]邊緣運算─實現普世人工智慧的最後一哩路 | 中研院訊。 (2021). 取自：https://newsletter.sinica.edu.tw/26211/(存取日期：2023年7月11日)
[16]什麼是 UDP？ | Cloudflare。 (2023). 取自：https://www.cloudflare.com/zh-tw/learning/ddos/glossary/user-datagram-protocol-udp/(存取日期：2023年7月11日)
[17]購買樹叢-肌電圖檢測器評價心得|台灣MakeHub。 (2018). 取自：https://ec.makehub.tw/seeed/101020058-grove-emg-detector(存取日期：2023年7月11日)
[18]【物聯網】入門介紹－樹莓派：一張信用卡大小的微電腦 | Jason Chen's Blog。 (2019). 取自：https://jason-chen-1992.weebly.com/home/introduction-to-raspberry-pi(存取日期：2023年7月11日)
[19]How To: PXE Boot your RPi. (2022). 取自：https://rob-ferguson.me/how-to-pxe-boot-your-rpi/(存取日期：2023年7月11日)
[20]Duckiebot DB18 人工智慧自走車 | Duckietown | 小鴨城 | 官方代理商 – RICELEE。 (2022). 取自：https://ricelee.com/product/official-duckiebot-db18-duckietown(存取日期：2023年7月11日)
[21]【QA】我們該如何給二分類模型好壞打分數？ | Cupoy。 (2021). 取自：https://www.cupoy.com/qa/club/ai_tw/0000016D6BA22D97000000016375706F795F72656C656173654B5741535354434C5542/0000017B7B4A6347000000236375706F795F72656C656173655155455354(存取日期：2023年7月11日)