Journal of Radio Electronics. eISSN 1684-1719. 2024. №3
Full text in Russian (pdf)
DOI: https://doi.org/10.30898/1684-1719.2024.3.8
A METHOD OF DATA SYNTHESIS
TO IMPROVE THE EFFECTIVENESS
OF NEURAL NETWORK TRAINING
D.A. Shlenskih 1, M.L. Belokopytov 1, D.V. Anohin 2, I.G. Ivanov3
1 Military space academy by A. Mozhaysky name
197198, Russia, St. Petersburg, Zhdanovskaya st. 132 St. Petersburg State University of Aerospace Instrumentation
190000, Russia, St. Petersburg, Bolshaya Morskaya st., 673 Main Test Spaces Center by G. Titov name
143090, Russia, Krasnoznamensk, Oktyabrskaya st. 3
The paper was received February 05, 2024
Abstract. The article is devoted to the formation of the initial data set for training neural networks. A proven method for creating synthetic data based on a graphics processor using a graphics pipeline is presented. A distinctive feature of this method is its modular architecture, which makes it easy to modify, delete or add individual stages to the synthetic image generation pipeline. A one-stage automatic detector based on a convolutional neural network of the Yolo type has been trained, the quality of the trained model and the operation of the recognition algorithm were also evaluated. Conclusions are drawn regarding the possibility of using this approach when creating representative samples of a large volume and their further use for training neural networks in pattern recognition.
Key words: neural network, training, synthetic data, graphics pipeline.
Corresponding author: Mark Lvovich Belokopytov, Hommer1990@mail.ru
References
1. Пчелинцев С.Ю., Юляшков М.А., Ковалева О.А. Метод создания синтетических наборов данных для обучения нейросетевых моделей распознаванию объектов [A method for creating synthetic datasets for training neural network models to recognize objects] // Информационно-управляющие системы. – 2022. – №. 3 (118). – С. 9-19. http://i-us.ru/index.php/ius/article/view/15242.
2. Сообщество IT-специалистов: Синтетические данные для машинного обучения: их природа, типы и способы генерации [Synthetic data for machine learning: their nature, types and methods of generation]: сайт. – Москва, 2019. – URL: https://habr.com/ru/articles/721170 (дата обращения: 21.11.2023). – Текст: электронный.
3. Гонахчян В.И. Модель производительности графического конвейера для однопроходной схемы рендеринга динамических трехмерных сцен [A graphics pipeline performance model for a single-pass rendering scheme for dynamic three-dimensional scenes] // Труды Института системного программирования РАН. – 2020. – Т. 32. – №. 4. – С. 53-72.
4. Леденёв М.А. Программный модуль повышения качества изображений в задачах распознавания [A software module for improving image quality in recognition tasks] // Известия Транссиба. – 2015. – №. 2 (22). – С. 110-115.
5. Малыхина Г.Ф., Меркушева А.В. Элементы статистической концепции обучения нейронной сети и прогнозирование точности ее функционирования [Elements of the statistical concept of neural network training and forecasting the accuracy of its functioning] // Научное приборостроение. – 2005. – Т. 15. – №. 1. – С. 29-45.
6. Belokopytov M.L., Shlenskikh D.A., Morozov S.V., Sirota S.V. Monitoring of maritime navigation facilities based on aerospace remote sensing data in the microwave range using neural network technologies. Zhurnal radioelektroniki [Journal of Radio Electronics] [online]. 2022. №4. https://doi.org/10.30898/1684-1719.2022.4.2 (In Russian).
7. Носырев П.В. и др. Валидация аналитических методик: теория и практика (Часть I) [Validation of analytical techniques: theory and practice (Part I)] // Ремедиум. Журнал о российском рынке лекарств и медицинской технике. – 2003. – №. 10. – С. 69-71.
For citation:
Shlenskih D.A., Belokopytov M.L., Anohin D.V., Ivanov I.G. A method of data synthesis to improve the effectiveness of neural network training. // Journal of Radio Electronics. – 2024. – №. 3. https://doi.org/10.30898/1684-1719.2024.3.8 (In Russian)