Алгоритмическое обеспечение нейросетевого обнаружения малоразмерных объектов в условиях высокой априорной неопределенности оптико-электронных данных
Работая с сайтом, я даю свое согласие на использование файлов cookie. Это необходимо для нормального функционирования сайта, показа целевой рекламы и анализа трафика. Статистика использования сайта обрабатывается системой Яндекс.Метрика
Научный журнал Моделирование, оптимизация и информационные технологииThe scientific journal Modeling, Optimization and Information Technology
Online media
issn 2310-6018

Algorithmic support for neural network detection of small objects in conditions of high a priori uncertainty of optoelectronic data

Kiryushin E.Y.,  Koshelev A.V. 

UDC 004.93
DOI: 10.26102/2310-6018/2026.62.11.006

  • Abstract
  • List of references
  • About authors

The article discusses a new algorithm for detecting small-size (point) targets in images of the mid-wavelength infrared range, operating under conditions of high a priori uncertainty of optoelectronic data and low signal-to-noise ratio. The proposed detection strategy is based on cascading multi-stage processing, which consistently reduces the uncertainty of the initial data. The algorithm includes a search for local maxima followed by threshold screening of deliberately false candidates based on amplitude characteristics, neighborhood approximation using the Gauss-Newton method, and the final classification of target-like objects using a compact four-layer convolutional neural network. At the same time, threshold filtering significantly reduces the number of false alarms even before applying computationally expensive neural network classification. To enable the integration of the algorithm into embedded software based on FPGA-based embedded software, static quantization of neural network parameters up to INT8 format has been applied. Experimental studies on real images of the mid-wave infrared range have confirmed the high efficiency of the developed approach: the model ensures the accuracy of detection of at least 0.95 with an integral quality metric of AUC = 0.96, and reducing the amount of memory for network weights by 3.14 times (up to 632 bytes) allows the algorithm to be used in resource-limited on-board real-time systems.

1. Wang Y., Gao K., Gebel E.S. End-To-End Mapless Navigation in Environments with Mixed Obstacles. In: 2026 12th International Conference on Automation, Robotics and Applications (ICARA), 05–07 February 2026, Istanbul, Turkiye. IEEE; 2026. P. 123–127. https://doi.org/10.1109/ICARA69401.2026.11480297

2. Kostrov V.V., Rakitin A.V., Bogatov A.D. Algorithmic support for the detection and coordination of point objects in optoelectronic systems with a bolometric matrix receiver. Radio Engineering and Telecommunication Systems. 2025;(4):44–54. (In Russ.). https://doi.org/10.24412/2221-2574-2025-4-44-54

3. Znamenskiy I.V. Dependence of the signal magnitude on displacement of the point spread function relative to the center of the photosensitive matrix pixel. Opticheskii Zhurnal. 2023;90(12):96–110. (In Russ.). https://doi.org/10.17586/1023-5086-2023-90-12-96-110

4. McKay M.B., Lawrence N.P., Wetton B., Gopaluni R.B. Error whitening: Why Gauss-Newton outperforms Newton. arXiv. URL: https://arxiv.org/abs/2605.11316 [Accessed 10th July 2026].

5. Fan M., Tian Sh., Liu K., Zhao J., Li Y. Infrared small target detection based on region proposal and CNN classifier. Signal, Image and Video Processing. 2021;15(4):1927–1936. https://doi.org/10.1007/s11760-021-01936-z

6. Moran J., Qing H. Brain-inspired filtering Network for small infrared target detection. Multimedia Tools and Applications. 2023;82(18):28405–28426. https://doi.org/10.1007/s11042-023-14762-x

7. Wei L., Ma Zh., Yang Ch., Yao Q. Advances in the Neural Network Quantization: A Comprehensive Review. Applied Sciences. 2024;14(17):7445. https://doi.org/10.3390/app14177445

8. Gholami A., Kim S., Dong Zh., Yao Zh., Mahoney M.W., Keutzer K. A Survey of Quantization Methods for Efficient Neural Network Inference. arXiv. URL: https://arxiv.org/abs/2103.13630 [Accessed 10th July 2026].

9. Sali S.M., Meribout M., Majeed A.A. Real Time FPGA Based CNNs for Detection, Classification, and Tracking in Autonomous Systems: State of the Art Designs and Optimizations. arXiv. URL: https://arxiv.org/abs/2509.04153 [Accessed 10th July 2026].

10. Hozhabr S.H., Giorgi R. A Survey on Real-Time Object Detection on FPGAs. IEEE Access. 2025;13:38195–38238. https://doi.org/10.1109/ACCESS.2025.3544515

Kiryushin Evgeniy Yurievich

Ural Optical and Mechanical Plant named after Mr. E.S. Yalamov

Ekaterinburg, Russian Federation

Koshelev Alexander Viktorovich

Peter the Great St. Petersburg Polytechnic University

Saint Petersburg, Russian Federation

Keywords: point target, infrared images, computer vision, cascade detection algorithm, gauss-Newton method, convolutional neural network, static quantization, INT8, embedded systems, FPGA

For citation: Kiryushin E.Y., Koshelev A.V. Algorithmic support for neural network detection of small objects in conditions of high a priori uncertainty of optoelectronic data. Modeling, Optimization and Information Technology. 2026;14(11). URL: https://moitvivt.ru/ru/journal/article?id=2597 DOI: 10.26102/2310-6018/2026.62.11.006 (In Russ).

© Kiryushin E.Y., Koshelev A.V. Статья опубликована на условиях лицензии Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NS 4.0)
111

Full text in PDF

Скачать JATS XML

Received 03.08.2026

Revised 19.10.2026

Accepted 11.11.2026