METHODS FOR IMPROVING THE ACCURACY OF OBJECT CLASSIFICATION WHEN BUILDING A VISION SYSTEM FOR EXPLORING CONFINED SPACES

M. A. Naumov, A. Y. Karpovsky

About the authors

M. A. Naumov, A. Y. Karpovsky

Карповский Артур Юрьевич – заведующий отделом автоматизации горных машин, Государственное бюджетное учреждение «Научно -исследовательский и проектно-конструкторский институт по автоматизации горных машин «Автоматгормаш им. В.А. Антипова», г. Донецк. Область научных интересов: приборостроение и радиоэлектроника, адрес 283003, г. Донецк, пр. Ильича, дом 93.

Download PDF

UDC: 519.4
DOI: 10.24412/2413-7383-2025-2-37-4-12
Language: Russian

Abstract: The article discusses methods for improving the accuracy of object classification in vision systems operating in confined spaces. Approaches to the synthesis of neural network models based on logical neurons and sigmoidal activation functions are proposed, providing adaptivity to incomplete and noisy data. The concept of fractal structure (FFS) and cooperative neural network architecture, combining private models to improve the accuracy of solutions, has been developed. The presented methods increase the interpretability and stability of neural networks, which is especially important for intelligent navigation and analysis systems in technically difficult conditions.

Keywords: Artificial neural networks, classification of objects, fractal structure of neural networks, vision systems, logical neurons.

References

1. Handbook of the theory of automatic control [Text] / A. G. Alexandrov, V. M. Artemyev, V. N. Afanasyev and others; edited by A. A. Krasovsky. — M.: Nauka, 1987. — 712 p. 2. Methods of classical and modern theory of automatic control [Text]. Vol
2. Statistical dynamics and identification of automatic control systems: textbook / K. A. Pupkov, N. D. Egupov, E. M. Voronov et al. edited by K. A. Pupkov, N. D. Egupov. — 2nd ed. revised and additional — M.: Publishing House of the Bauman Moscow State Technical University, 2004. — 640 p.
3. Methods of classical and modern theory of automatic control [Text]. Vol. 5. Methods of modern theory of automatic control: textbook / K. A. Pupkov, N. D. Egupov, A. I. Barkin et al. edited by K. A. Pupkov, N. D. Egupov. — 2nd ed. revised and additional — M.: Publishing House of the Bauman Moscow State Technical University, 2004. 784 p.
4. Methods of robust, neuro-fuzzy and adaptive control [Text]: textbook / K. A. Pupkov, N. D. Egupov, A. I. Gavrilov et al. edited by N. D. Egupov; ed. 2nd, stereotypical. Moscow: Publishing House of the Bauman Moscow State Technical University, 2002. 744 p.
5. Kolmogorov, A. N. On the representation of continuous functions of several variables in the form of a superposition of continuous functions of one variable and addition [Text] / A. N. Kolmogorov // Dokl. USSR Academy of Sciences. 1957. Vol. 114, No. 5. pp. 953-956.
6. Hecht-Nielsen R. Kolmogorov's mapping neural network theorem [Text] // IEEE standards first Annual Int. Conference on Neural Networks. — 1987. — Volume 3. —pp. 11-13.
7. Yurkov, D. A. Method of synthesis of neural network model structures [Text] / D. A. Yurkov // Visn. Skydnoukr. national un-tu. — 2009. — № 5 (135). — Pp. 115-122.
8. Yurkov, D. A. Synthesis of neural networks with fractal structure [Text] / D. A. Yurkov // East European Journal of Advanced Technologies. — 2009. — № 4-3 (40). — Pp. 39-44.
9. Yurkov, D. A. Classification based on cooperative neural network structures [Text] / D. A. Yurkov // East European Journal of Advanced Technologies. — 2009. — №5-3 (41). — Pp. 51-56.
10. Sivak, M. A. A study of the applicability of robust loss functions in neural networks [Text] / M. A. Sivak // Collection of scientific papers of NSTU. - 2020. — No. 4. — pp. 50-58.
11. Sivak, M. A. Optimization of the operation of a robust neural network for the classification problem [Text] / M. A. Sivak, V. S. Timofeev // Science. Technologies. Innovations: collection of scientific papers: at 9 a.m. Novosibirsk, November 30 - December 4, 2020, Part 2. NSTU Publishing House, 2020. pp. 298-300.
12. Gudzimo, S. Dynamically developing search process [Text] / S. Gudzimo, D. Meger, J. Purchase. — 2020. — URL: https://inlnk.ru/poZ0mB.
13. Sivak, M. A. Classification of noisy data with different sample sizes [Text] / M. A. Sivak // Science. Technologies. Innovations: collection of scientific papers: at 9 a.m. Novosibirsk, December 6-10, 2021, Part 2. — NSTU Publishing House, 2021. — pp. 277-279.
14. Sivak, M. A. Setting up robust neural networks to solve the classification problem [Text] / M. A. Sivak, V. S. Timofeev // Reports of Tomsk State University of Control Systems and Radio Electronics. — 2021. — Vol. 24, No. 3. — pp. 26-32.
15. Sivak, M. A. Construction of robust neural networks with various loss functions [Text] / M. A. Sivak, V. S. Timofeev // Data analysis and processing systems. — 2021. — Vol. 82, No. 2. — pp. 67-83.
16. Timofeev, V. S. Robust neural network with simple architecture [Text] / V. S. Timofeev, M. A. Sivak // Siberian Journal of Industrial Mathematics. — 2021. — Vol. 24, No. 4. — pp. 126-138.
17. Yukang Chen, Yanwei Li, Tao Kong, Lu Qi, Ruihang Chu, Lei Li, Jiayaya Jia. Mathematical exhortation based on user needs // CVPR. 2021. pp. 9563-9572.
18. Rowel Atienza. Fascination with the volume of data for understanding the text of a book // ICCVW. 2021. pp. 1561-1570.
19. Zhong Zhong, Liang Zheng, Guoliang Kang, Shaozi Li, Yi Yang. Fascination with the volume of data under random circumstances // AAAI. 2020. pp. 13001-13008.
20. Ivanov D. A. Turuta D. V. Application of data augmentation methods to improve the accuracy of image classification // Artificial intelligence and decision—making. - 2022. — No. 3. — pp. 45-56.
21. Talukdar S. Mahato S. Shahfahad Pal S. Liu Ya. Rahman A. Classification of vegetation cover in the field of land use using machine learning classifiers for satellite observations - an overview. Updated Sens

Issue: 2(37)'2025
Section: ARTIFICIAL INTELLIGENCE AND MACHINE LEARNING
How to cite: M. A. Naumov, A. Y. Karpovsky. METHODS FOR IMPROVING THE ACCURACY OF OBJECT CLASSIFICATION WHEN BUILDING A VISION SYSTEM FOR EXPLORING CONFINED SPACES // Problems of Artificial Intelligence. - 2025. - № 2 (37). - P. 4-12. - https://paijournal.guiaidn.ru/en/2025/2(37)-1.html

Back to issue 2(37)'2025