V.B. Suchkov1, A.Y. Perov2, G.L. Pavlov3, S.G. Tseytlin4

1–4 Bauman Moscow State Technical University (Moscow, Russia)
1 vbs-2014@bmstu.ru, 2 perovau@bmstu.ru, 3 pavlov_503@bmstu.ru, 4 tseytlinsg@student.bmstu.ru

Statement of the problem. The automatic recognition of radar targets based on range profiles is limited by redundancy, noisy data, and incomplete use of polarimetric information in existing approaches. This reduction in classification reliability can be attributed to these factors. In order to resolve this issue, it is imperative to establish a representative training sample based on multi-point target models with additive Gaussian noise and undertake a comparative analysis of the efficacy of neural network models for six categories of input data: polarisation scattering matrix, Pauli and Friedman-Durdin decompositions, their concatenations, as well as a single-channel range portrait. The comparison criterion encompasses not only classical quality metrics (e.g. accuracy, precision, completeness), but also the visualisation of the structure of feature spaces formed by the neural network using the t-SNE method. This allows for the evaluation of the separability of classes at a deep level of data representation.

Purpose. The objective of this study is to assess the impact of additional features extracted from Pauli and Freeman–Derden polarimetric decompositions on the effectiveness of solving the problem of radar object classification based on a one-dimensional convolutional network supplemented with a cross-channel attention mechanism for adaptive weighting of feature informativeness.

Results. The effectiveness of polarimetric data for training neural network classifiers is demonstrated. Using range profiles without polarimetry degrades accuracy to 84.5% compared to the polarimetric scattering matrix baseline achieving 98.9% accuracy. Standalone Pauli or Freeman–Durden decompositions reduce classification accuracy (95.0%/71.8%), however concatenation with polarimetric scattering matrix yields improvements (+0.7%/+0.2%). The best performance is achieved with PSM and Pauli components (99.6%); t-SNE confirms improved class separability. Trajectory flight tests also demonstrate polarimetric data superiority.

Practical significance. The proposed architecture ensures high classification accuracy with minimal computational costs. The complementarity of polarimetric scattering matrix and Pauli decomposition is established as an effective augmentation for robust automatic target recognition under real-world conditions.

Suchkov V.B., Perov A.Y., Pavlov G.L., Tseytlin S.G. Technique of selection of targets based on neural network processing of polarized radar range profiles. Achieve­ments of modern radioelectronics. 2026. V. 80. № 4. P. 35–53. DOI: https://doi.org/10.18127/j20700784-202604-05 [in Russian]

1. Soleti R. & Cantini L. & Berizzi F. & Capria A. & Calugi D. Neural network for polarimetric radar target classification. European Signal Processing Conference. 2006.
2. Gong J., Xiang Y., Zhang Q. Synthetic Aperture Radar Target Recognition with Limited Training Data Based on 3D Scattering Center Model. Remote Sensing. 2021. V. 13. № 15. P. 2939. DOI: 10.3390/rs13152939
3. Ge J. i et al. A Physically Interpretable Rice Field Extraction Model for PolSAR Imagery. Remote Sens. MPDI. 2023. R. 50
4. Mete Ahishali, Serkan Kiranyaz, Turker Ince & Moncef Gabbouj. Classification of polarimetric SAR images using compact convolutional neural networks, GIScience & Remote Sensing. 2021. 58:1. 28-47. DOI: 10.1080/15481603.2020.1853948
5. Barrachina J., Ren C., Morisseau C., Vieillard G., Ovarlez J.-Ph. Complex-valued neural networks for polarimetric sar segmentation using pauli representation. IEEEIGARSS 2022. Jul 2022. Kuala Lumpur. Malaysia. ff10.1109/IGARSS46834.2022.9883251ff. ffhal03752893
6. Suchkov V.B., Perov A.Yu. Algoritm polyarizacionnoj selekcii ob``ektov v radiolokatore s inversny`m sintezirovaniem apertury` na osnove nejroseti Squeezenet. Doklady` XXVI Mezhdunar. konf. «Cifrovaya obrabotka signalov i ee primenenie» (DSPA – 2024: Moskva, 27–29 marta 2024 g.). M.: Rossijskoe nauchno-texnicheskoe obshhestvo radiotexniki, e`lektroniki i svyazi im. A.S. Popova. 2024. S. 119–123. EDN BFUMHC.
7. Huan S., et al. «Radar Human Activity Recognition with an Attention-Based Lightweight Network» Sensors. PMC. 2023.
8. Suchkov V.B., Perov A.Yu., Ivanov V.M. Issledovanie polyarizacionnoj selekcii radiolokacionny`x ob``ektov v X-diapazone radiovoln na osnove svertochny`x nejronny`x setej razlichnoj arxitektury`. Sb. trudov XXXI Mezhdunar. nauchno-texn. konf. «Radiolokaciya, navigaciya, svyaz`» (Voronezh, 15–17 aprelya 2025 g.). V 6 tomax. Voronezh: Voronezhskij gosudarstvenny`j universitet. 2025. S. 83–92. EDN MEDALS.
9. Fu Zhequan & Li Junrong & Li Xiangping & Dan Bo & Wang Xukun. A Neural Network with Convolutional Module and Residual Structure for Radar Target Recognition Based on High-Resolution Range Profile. Sensors. 2020. 20. 586. 10.3390/s20030586
10. Wang Q., Wu B., Zhu P., Li P., Zuo W. and Hu Q. ECA-Net: Efficient channel attention for deep convolutional neural networks. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition (Piscataway, NJ: IEEE). 2020. 11534–11542.
11. Yang W., et al. Dual-band polarimetric HRRP recognition via a brain-inspired multi-channel fusion feature extraction network (MFFE-Net). Frontiers in Neuroscience. 2023.
12. Muttenthaler L., Hollenstein N. and Barrett M. Human brain activity for machine attention. arXiv [Preprint]. 2020. doi: 10.48550/arXiv.2006.05113
13. Ly`senko P.V., Nasonov I.A., Galyaev A.A., Berlin L.M. Nejrosetevoj podxod k klassifikacii akusticheskix signalov s ispol`zovaniem informacionny`x priznakov. Dokl. RAN. Matem., inform., procz. upr., 514:2 (2023), 39–48; Dokl. Math., 108:suppl. 2 (2023), S196–S204.
14. Kumar A., Giusti E. and Martorella M. Hybrid polarimetry inverse synthetic aperture radar. IET Radar, Sonar & Navigation. 2025. 19(1) P. e70004.
15. Lixoedenko K.P., Seregin G.M., Suchkov V.B., Perov A.Yu. Matematicheskoe modelirovanie polyarizacionny`x radiolokacionny`x portretov ob``ektov slozhnoj formy` na osnove ix mnogotochechny`x modelej v radiolokatorax s inversny`m sintezirovaniem apertury`. Uspexi sovremennoj radioe`lektroniki. 2024. T. 78. № 2. S. 13–24. DOI 10.18127/j20700784-202402-02
16. Lee J.-S., & Pottier E. Polarimetric Radar Imaging: From Basics to Applications (1st ed.). CRC Press. 2009. https://doi.org/10.1201/ 9781420054989