Journal of Radio Electronics. eISSN 1684-1719. 2025. №11

Contents

Full text in Russian (pdf)

Russian page

 

 

DOI: https://doi.org/10.30898/1684-1719.2025.11.34

 

 

 

THE STUDY OF BINARY CLASSIFIERS

BASED ON DIFFERENT MATHEMATICAL APPROACHES

WHEN PERFORMING TASKS OF RECOGNIZING MOTOR IMAGES

IN AN ASYNCHRONOUS BRAIN-COMPUTER INTERFACE

 

D.V. Zhuravlev ¹, A.A. Reznichenko ¹, A.N. Golubinsky ¹, A.A. Tolstykh ²

 

¹ Voronezh State Technical University, 394006, Voronezh, 20th Anniversary of October St., 84

² LTD «RTK», 107140, Moscow, Verkhnyaya Krasnoselskaya st., 16с1

 

The paper was received September 11, 2025.

 

Abstract. Currently, many attempts are being made to create an asynchronous brain-computer interface that could recognize motor patterns with high accuracy. However, in practice, achieving recognition accuracy of even more than 60 % online is a difficult task. Especially when using portable equipment. In this paper, an attempt is made to create a portable small-sized asynchronous brain-computer interface designed for real-time recognition of motor imagery with a binary classification accuracy of at least 65 %. To achieve this goal, a mock-up of equipment was created that transmits online recorded electroencephalographic signals to a personal computer via a Wi-Fi radio channel. The standalone model of the equipment is placed on a portable neural headset with resistive-capacitive electrodes of the “dry” type. The paper considers the theoretical aspects of the approaches used. A block diagram of a software system implementing motor imagery recognition is presented. The results of the experimental application of the proposed system are analyzed. Software has been developed in Python for signal preprocessing, feature extraction, and classification. As part of the work, several classifiers based on different mathematical approaches were studied: linear discriminant analysis (LDA), multilayer perceptron, convolutional neural network, which is a repeating part of the architecture that consists of a set of different layers (ResNet). The architecture of the network consisting of eighteen layers has been developed. Optimization methods of the first and second orders were also investigated. The evaluation of the influence of optimization algorithms on the work of neural network classifiers was carried out using the metrics Accuracy, Precision, Recall and F1 measure. In total, four optimization algorithms were studied: the adaptive impulse optimization algorithm (ADAM), the Levenberg-Marquardt algorithm (LM), the stochastic gradient descent algorithm (SGD), and the Broyden-Fletcher-Goldfarb-Shanno algorithm (BFGS). The modernization of the Levenberg-Marquardt algorithm is proposed in relation to the tasks of classifying motor imagery. The highest classification accuracy (Accuracy metric) of motor imagery was achieved using a classifier based on ResNet and the ADAM optimization algorithm. The classification accuracy in offline mode was 68.31 % (Accuracy metric) and 68.80 % (Recall metric), while in online mode it was 65.92 % and 66.20 %, respectively.

Key words: brain-computer interface, motor images, gradient descent, Levenberg-Marquardt algorithm, convolutional neural network, ResNet, linear discriminant analysis, fully connected perceptron, BFGS optimization algorithm.

Financing: The research was carried out at the expense of a grant from the Russian Science Foundation No. 24-29-20168, https://rscf.ru/project/24-29-20168/.

Corresponding author: Zhuravlev Dmitry Vladimirovich, ddom1@yandex.ru

 

References

1. Zhuravlev D.V. Alphanumeric information transmission system based on time-frequency analysis of the electroencephalogram signal. // Journal of Radio Electronics. – 2025. – №. 7. https://doi.org/10.30898/1684-1719.2025.7.15 (In Russian)

2. Suryotrisongko H., Samopa F. Evaluating openbci spiderclaw v1 headwear's electrodes placements for brain-computer interface (BCI) motor imagery application // Procedia Computer Science. – 2015. – Т. 72. – С. 398-405.

3. Kapralov N.V., Nagornova Z.V., Shemyakina N.V. Classification methods for EEG patterns of imaginary movements // Informatics and Automation. – 2021.  – Т. 20. – №. 1. – С. 94-132.

4. Zhuravskaya A., Stankevich L.A. Primenenie neinvazivnogo interfejsa «mozg-komp`yuter» dlya klassifikacii voobrazhaemy`x dvizhenij nizhnix konechnostej cheloveka // Sistemny`j analiz v proektirovanii i upravlenii. – 2021. – T. 25.  – №. 3. – S. 146-158.

5. Pavlenko D.V., Tataris Sh. E`., Ovcharenko V.V. Primenenie glubokogo obucheniya v interfejsax mozg–komp`yuter dlya raspoznavaniya dvizhenij // Programmny`e produkty` i sistemy`. – 2024. – T. 37. – №. 2. – S. 164-169.

6. Echtioui A. et al. A novel convolutional neural network classification approach of motor-imagery EEG recording based on deep learning // Applied Sciences.  – 2021. – Т. 11. – №. 21. – С. 9948.

7. Runnova A.E. i dr. Klassifikaciya patternov dvigatel`noj aktivnosti na E`E`G-danny`x // Vestnik rossijskix universitetov. Matematika. – 2017. – T. 22. – №. 5-2. – S. 1127-1132.

8. Asadullaev R.G., Afonin A.N., Shhetinina E.S. Raspoznavanie patternov dvigatel`noj aktivnosti nejronnoj set`yu po neprery`vny`m danny`m opticheskoj tomografii fNIRS // E`konomika. Informatika. – 2021. – T. 48. – №. 4. – S. 735-746.

9. The official website of the project «OpenBCI». URL: https://github.com/OpenBCI/ (accessed: 01.09.2025).

10. Zhuravlev D.V., Golubinskij A.N., Tolsty`x A.A. Razrabotka metodiki nastrojki parametrov interfejsov «mozg-komp`yuter» dlya provedeniya e`ksperimentov po klassifikacii motorny`x obrazov v programme OpenVibe // Biomedicinskaya radioe`lektronika. – 2025. – T. 28. – № 3. – S. 15-30.

11. Gudfellou Ya., Ioshua B., Kurvill` A. Glubokoe obuchenie. – Litres, 2017.

12. Shafiq M., Gu Z. Deep residual learning for image recognition: A survey // Applied sciences. – 2022. – Т. 12. – №. 18. – С. 8972.

13. Heuts S. et al. Bayesian analytical methods in cardiovascular clinical trials: why, when, and how // Canadian Journal of Cardiology. – 2025. – Т. 41. – №. 1. – С. 30-44.

14. Li Q., Shao J. Sparse quadratic discriminant analysis for high dimensional data // Statistica Sinica. – 2015. – С. 457-473.

15. Xajkin S. Nejronny`e seti: polny`j kurs, Izdatel`skij dom «Vil`yams» // Moskva. – 2006. – T. 1. – S. 104.

16. Bure V.M., Parilina E.M., Svirkin M.V. Teoriya veroyatnostej. – 2008.

17. Miconi T. Hebbian learning with gradients: Hebbian convolutional neural networks with modern deep learning frameworks // arXiv preprint arXiv:2107.01729. – 2021.

18. Siri B. et al. A mathematical analysis of the effects of Hebbian learning rules on the dynamics and structure of discrete-time random recurrent neural networks // Neural computation. – 2008. – Т. 20. – №. 12. – С. 2937-2966.

19. Nawi N.M., Ransing M.R., Ransing R.S. An improved learning algorithm based on the Broyden-Fletcher-Goldfarb-Shanno (BFGS) method for back propagation neural networks // Sixth International Conference on Intelligent Systems Design and Applications. – IEEE, 2006. – Т. 1. – С. 152-157.

20. Ashby W. Design for a brain: The origin of adaptive behaviour. – Springer Science & Business Media, 2013.

21. Ji L. et al. An intelligent diagnostic method of ECG signal based on Markov transition field and a ResNet // Computer Methods and Programs in Biomedicine. – 2023. – Т. 242. – С. 107784.

22. Simonyan K., Zisserman A. Very deep convolutional networks for large-scale image recognition // arXiv preprint arXiv:1409.1556.  – 2014.

23. Krizhevsky A., Sutskever I., Hinton G.E. Imagenet classification with deep convolutional neural networks // Advances in neural information processing systems. – 2012.  – Т. 25.

24. Ioffe S., Szegedy C. Batch normalization: Accelerating deep network training by reducing internal covariate shift // International conference on machine learning.  – pmlr, 2015. – С. 448-456.

25. Nguyen N., Han S. AET-SGD: Asynchronous Event-triggered Stochastic Gradient Descent // arXiv preprint arXiv:2112.13935.  – 2021.

26. Kingma D.P. Adam: A method for stochastic optimization // arXiv preprint arXiv:1412.6980.  – 2014.

27. Going deeper with convolutions | IEEE Conference Publication | IEEE Xplore. URL: https://ieeexplore.ieee.org/document/7298594 (accessed: 10.09.2025).

28. Shepherd A.J. Second-order methods for neural networks: Fast and reliable training methods for multi-layer perceptrons. – Springer Science & Business Media, 2012.

29. Smith J.S., Wu B., Wilamowski B.M. Neural network training with Levenberg–Marquardt and adaptable weight compression // IEEE transactions on neural networks and learning systems. – 2018. – Т. 30. – №. 2. – С. 580-587.

30. Golubinskij A.N., Tolsty`x A.A. Gibridny`j metod obucheniya svertochny`x nejronny`x setej // Informatika i avtomatizaciya. – 2021. – T. 20. – №. 2. – S. 463-490.

31. PyTorch. URL: https://pytorch.org/ (accessed: 10.09.2025).

For citation:

Zhuravlev D.V., Reznicenko A.A., Golubinsky A.N., Tolstykh A.A. The study of binary classifiers based on different mathematical approaches when performing tasks of recognizing motor images in an asynchronous brain-computer interface // Journal of Radio Electronics. – 2025. – №. 11. https://doi.org/10.30898/1684-1719.2025.11.34