P.A. Kurakin1, A.N. Korotkikh2, A.S. Komogortsev3, V.S. Chernov4, A.B. Gerasimov5, V.V. Mikhailov6

1–4,6 JSC «Rybinsk Instrument Engineering Plant» (Rybinsk, Russia)

5 P.G. Demidov Yaroslavl State University (Yaroslavl, Russia)

1 yepseeke.kaps@yandex.ru, 2 ntc@rzp.su, 3 pribor@rzp.su, 4 vladchernov77@mail.ru, 5 gerasimov@uniyar.ac.ru, 6 ntc@rzp.su

Ensuring the safety of flights of unmanned aerial vehicles (UAVs), as well as preventing their illegal use, requires monitoring air traffic at sufficiently large distances, at any time of the day, including when flying in radio silence. The specified requirements are best met by monitoring using radar stations (radars). At the same time, along with detecting and measuring the UAV movement parameters, it is necessary to perform UAV type recognition in order to compare with the flight mission or determine the degree of threat. In addition, along with UAVs, radars can detect flying birds, people, motor vehicles and water transport during operation. In this regard, it is necessary to distinguish UAVs from many other detectable objects.

The goal of this work is to ensure the recognition of objects detected by the Cheremukha multifunctional radar and the selection of UAVs based on the analysis of the Microdopler structure of echo signals using an convolutional neural network.

Based on the results of the work carried out, it was possible to achieve good quality training for the DenseNet 121 neural network in recognizing 8 classes of objects, including small and large multirotor-type UAVs, based on the analysis of Microdopler modulation of the echo signal. The quality of recognition does not significantly depend on the length of the interval for constructing mel spectrograms. In this regard, DenseNet 121 variant is being introduced into the Cheremukha radar, trained with a duration of 0,2 s. The average precision of UAV recognition in this variant is 0,7588, and the recall is 0,9545.

Kurakin P.A., Korotkikh A.N., Komogortsev A.S., Chernov V.S., Gerasimov A.B., Mikhailov V.V. Object recognition based on a convolutional neural network. Achievements of modern radioelectronics. 2025. V. 79. № 10. P. 16–22. DOI: https://doi.org/10.18127/j20700784-202510-03 [in Russian]

1. Molchanov P. et al. Classification of small UAVs and birds by micro-Doppler signatures. International Journal of Microwave and Wireless Technologies. 2014. T. 6. № 3-4. S. 435–444.
2. Torvik B., Olsen K. E., Griffiths H. Classification of birds and UAVs based on radar polarimetry. IEEE geoscience and remote sensing letters. 2016. T. 13. №. 9. S. 1305–1309.
3. Derkachev V.A., Bakhchevnikov V.V., Bakumenko A.N. Klassifikator BPLA mul'tirotornogo tipa. Izvestiya YuFU. Tekhnicheskie nauki. 2023. № 2. S. 90–99. [in Russian]
4. Sverdlov B.G., Igonin D.M., Vylegzhanin I.S., Kutina A.V., Pushkov A.A. Radiolokatsionnoe raspoznavanie BPLA i ornitologicheskikh ob"ektov (ptits). Vestnik vozdushno-kosmicheskoy oborony. 2019. № 4. S. 41–50. [in Russian]
5. Mendis G. J., Wei J., Madanayake A. Deep learning cognitive radar for micro UAS detection and classification. 2017 Cognitive Communications for Aerospace Applications Workshop (CCAA). IEEE, 2017. S. 1–5.
6. Raval D. et al. Convolutional neural networks for classification of drones using radars. Drones. 2021. T. 5. №. 4. S. 149.