Abstract. Comparison of the effectiveness of the convolutional neural network (CNN) and radial basis func-tions (RBFs) algorithms in the processing of forearm electromyograms (EMG) obtained over several days is performed. Pre-processed signal is used as the input data, the CNN algorithm is applied as the input feature vector for the raw data. The both methods are shown to have several perspectives for the use in device control tasks. The CNN is able to recognize the characteristic signs of EMG without prelimi-nary signal processing, despite their stochastic nature. This allows to provide feature selection or trans-formation of the signal before classification. The method of data classification with pre-processing demonstrates greater recognition accuracy for RBF. A pronounced dependence of the recognition accu-racy on the subject and on the time between obtaining samples was obtained. Thus, although the SNA algorithm showed good results for the raw EMG signal, the training sample with multi-day data allowed to achieve stable performance, which imposes limitations on its use in clinical systems.

Keywords: electromyogram, prosthesis, biocontrol, human-machine interface, machine learning, artificial neural networks

References:

1. Biron K., Englehart K. Proceedings of the 18th Congress of the International Society of Electrophys-iology and Kinesiology, Aalborg, Denmark, 16–19 June 2010.
2. He J., Zhang D., Jiang N., Sheng X., Farina D. J. Neural Eng., 2015, no. 12.
3. Zia ur Rehman M., Gilani S., Waris A., Niazi I., Slabaugh G., Farina D., Kamavuako E. Appl. Sci., 2018, no. 7(8), DOI: 10.3390/app8071126.
4. Rehman Z.U. et al. Sensors, 2018, no. 8(18), DOI: 10.3390/s18082497.
5. Lee S., Kruse J. Analog Devices, 2008, no. 200, pp. 1–2.
6. Mendez I., Hansen B.W., Grabow C.M., Smedegaard E.J., Skogberg N.B., Uth X.J., Bruhn A., Geng B., Kamavuako E.N. Proceedings of the 2017 International Conference on Rehabilitation Robotics (ICORR), London, UK, 17–20 July 2017, pp. 1211–1214.
7. Pizzolato S., Tagliapietra L., Cognolato M., Reggiani M., Müller H., Atzori M. PLoS ONE, 2017, no. 12.
8. Benatti S., Casamassima F., Milosevic B., Farella E., Schönle P., Fateh S., Burger T., Huang Q., Benini L. IEEE Trans. Biomed. Circuits Syst., 2015, no. 9, pp. 620–630.
9. Amirabdollahian F., Walters M. Proceedings of the International Conference on Rehabilitation Ro-botics (ICORR2017), London, UK, 17–21 July 2017.
10. Montoya M., Henao O., Munoz J. Proceedings of the XVIII International Conference on Human Computer Interaction, Cancun, Mexico, 25–27 September 2017.
11. Masson S., Fortuna F., Moura F., Soriano D. Proceedings of the XXV Congresso Brasileiro de Engenharia Biomédica, Rio de Janeiro, Brazil, 17–20 October 2016.
12. Boyali A., Hashimoto N., Matsumoto O. Proceedings of the 2015 IEEE 4th Global Conference on Consumer Electronics (GCCE), Osaka, Japan, 27–30 October 2015, pp. 200–201.
13. Budko R.Yu., Chernov N.N. Modeling, Optimization and Information Technology, 2019, no. 1 (7), pp. 54–66.
14. Patent RU2635632C1, Sposob i sistema upravleniya intellektual'noy bionicheskoy konechnost'yu (Method and Control System of Intellectual Bionic Limb), Ivanyuk N.M., Karimov V.R., Budko R.Yu., Gronskiy P.V., Kleyman S.M., Published 11.14.2017.
15. Budko R., Starchenko I., Budko A. Lecture Notes in Computer Science, 2016, no. 9812, pp. 163–171.
16. Budko R.Yu., Starchenko I.B. Trudy SPIIRAN (SPIIRAS Proceedings), 2016, no. 46, pp. 76–89. (in Russ.)
17. Atzori M., Cognolato M., Müller H. Front. Neurorobot., 2016, vol. 10, art. 9. DOI: 10.3389/fnbot.2016.00009.
18. Saichon J., Chidchanok L., Suphakant P. IEEE Trans Neural Netw., 2010, no. 3(21), pp. 381–392.