-
Radar Cross Section (RCS), a crucial physical quantity that characterizes the backscattering intensity of targets under radar illumination, is the primary metric for assessing stealth capabilities. With the progression of radar detection technologies, RCS reduction has become a forefront research topic in radar stealth, aiming to minimize target detectability. As radar networking technologies mature, Bistatic Radar RCS reduction is gaining increasing significance over Monostatic Radar RCS reduction in future electromagnetic stealth countermeasures. Artificial electromagnetic metasurface have introduced innovative technical pathways for Bistatic Radar RCS reduction. However, current metasurface designs still face challenges related to inefficiency and suboptimal performance, primarily due to the time-consuming nature of large-scale array optimization and the global extremum characteristics of Bistatic Radar RCS reduction. To overcome these limitations, this study proposes a Few-Shot Convolutional Neural Network (CNN)-based approach, which achieves uniform full-space radar echo scattering by directionally optimizing metasurface phase distributions, thereby enabling effective Bistatic Radar RCS reduction. The approach integrates convolutional feature extraction, residual enhancement, and fully connected optimization modules, alongside a customized loss function, to efficiently capture the complex multidimensional relationships between diffuse reflection phases and the full-space RCS extrema. Theoretical calculations, full-wave simulations, and experimental tests show that the metasurface designed with this approach achieve over 10 dB of Bistatic Radar RCS reduction within the 7.26 GHz to 10.74 GHz frequency range. The method also ensures uniform diffuse reflection across the full space for various incidence angles (30°、45°、60°). Compared to traditional optimization algorithms, this method enhances RCS reduction by 17.2% while significantly improving computational efficiency. This approach offers a promising new technical paradigm for achieving full-space electromagnetic stealth in advanced weapon systems.
-
图 1 (a) 随机相位分布; (b) 随机相位 3D 远场图; (c) 随机相位 2D 远场图
Figure 1. (a) Random phase distribution; (b) 3D far-field of random phase; (c) 2D far-field of random phase
图 2 (a)—(d) GA, PSO, PSO-GA, PSO-SA优化后的相位分布; (e)—(h) 优化相位分布后的 3D 远场图; (i)—(l) 优化相位分布后的 2D 远场图
Figure 2. (a)–(d) Phase distributions optimized by GA, PSO, PSO-GA, and PSO-SA; (e)–(h) 3D far-field patterns of the optimized phase distributions; (i)–(l) 2D far-field patterns of the optimized phase distributions
图 3 优化算法迭代过程中 RCS 均值的变化趋势
Figure 3. The Variation Trend of RCS Mean Value in the Iteration Process of Optimization Algorithms
图 4 CNN 模型示意图
Figure 4. CNN Model Diagram
图 5 全连接层结构示意图
Figure 5. Diagram of Fully Connected Layer Structure
图 6 (a) Leaky ReLU激活函数图; (b) Sigmoid激活函数图
Figure 6. (a) Leaky ReLU Activation Function Graph; (b) Sigmoid Activation Function Graph
图 7 (a) 损失函数图; (b) RCS 值迭代曲线图; (c) 模型运行 100 次结果图
Figure 7. (a) Loss Function Diagram; (b) RCS Value Iteration Curve Diagram; (c) Results of 100 Runs with Model
图 8 (a) CNN 优化后的相位分布; (b) CNN 优化相位分布后的 3D 远场图; (c) CNN 优化相位分布后的 2D 远场图
Figure 8. (a) Phase distribution optimized by CNN; (b) 3D far-field patterns after optimizing phase distribution by CNN; (c) 2D far-field patterns after optimizing phase distribution by CNN
图 9 (a) 各类算法优化效果; (b) 各类算法运行时间
Figure 9. (a) Optimization effect of various algorithms; (b) Running time of various algorithms
图 10 (a) 单元图; (b) 2 × 2子阵图
Figure 10. (a) Cell diagram; (b) 2 × 2 submatrix
图 11 (a)—(d) PEC, 随机, PSO-SA 优化, CNN 优化相位分布后的 3D 远场图; (e)—(h) PEC, 随机, PSO-GA 优化, CNN优化相位分布后的 2D 远场图
Figure 11. (a)–(d) 3D far-field patterns after phase distribution optimization by PEC, Random, PSO-SA, and CNN, respectively; (e)–(h) 2D far-field plots after phase distribution optimization by PEC, Random, PSO-SA, and CNN respectively
图 12 入射角分别为 30°、45°、60° : (a)—(c) CST 全波仿真 PEC 板的 3D 远场图; (d)-(f) CNN 优化后的相位分布; (g)—(i) CNN 优化后的相位分布在 CST 全波仿真的 3D 远场图; (j)—(l) CNN 优化后的相位分布在 CST 全波仿真的 2D 远场图
Figure 12. Incident angles of 30°, 45°, 60°: (a)–(c) 3D far-field patterns of CST full-wave simulation of PEC plates; (d)–(f) Phase distribution optimized by CNN; (g)–(i) 3D far-field patterns of CST full-wave simulation with CNN-optimized phase distribution; (j)–(l) 2D far-field patterns of CST full-wave simulation with CNN-optimized phase distribution
图 13 (a) CST 全波仿真中 RCS 值随频率变化曲线; (b) 双站 RCS 减缩值
Figure 13. (a) RCS vs Frequency Curve in CST Full-Wave Simulation; (b)Bistatic RCS reduction
图 14 (a)超表面加工样品示意图; (b) 暗室测试环境
Figure 14. (a) Schematic Illustration of the Metasurface Fabrication Sample; (b) Darkroom Testing Environment
图 15 (a)—(b) PEC 板全波仿真与实测 1D 远场结果图; (c)—(d) 样品全波仿真与实测 1D 远场结果图
Figure 15. (a)–(b) 1D Far-field Results for Full-wave Simulation and Measurement of PEC; (c)–(d) 1D Far-field Results for Full-wave Simulation and Measurement of Sample
图 16 双站 RCS 减缩仿真与实测值
Figure 16. Bistatic RCS Reduction Simulation and Measured Values
表 1 损失函数不同权重参数效果对比
Table 1. The comparison of the effects of different weight coefficients of the loss function.
权重参数 参数取值 RCS 值 $ \gamma_{RCS} $ 0.1, 0.5, 1.0, 1.5 15.5, 14.7, 14.6, 14.6 (收敛速度慢) $ \gamma_{Phase} $ 0.1, 0.3, 0.7, 1.0 16.8, 16.3, 16.2, 16.0 (无法减缩 RCS) $ \gamma_{RCS} $ + $ \gamma_{Phase} $ (0.1, 0.1), (0.5, 0.1), (0.5, 0.3)... 16.3, 14.8, 15.6... (二值化模糊) $ \gamma_{RCS} $ + $ \gamma_{Phase} $ + $ \gamma_{Reg} $ (0.5, 0.1, 0.5), (0.5, 0.1, 1), (0.5, 0.1, 1.5)... 14.8, 14.7, 14.6... 
DownLoad: CSV
表 A1 实验环境配置
Table A1. Experimental environment configuration.
名称 配置信息 开发语言 Python 3.9 框架 PyTorch 1.10.0 + CUDA 12.0 CPU Intel Core i9 GPU GeForce RTX 4060 Laptop GPU (8G) 内存 8G NumPy 1.21.3 Matplotlib 3.9.2 torchvision 0.13.0 Pandas 1.3.3
DownLoad: CSV
-
[1] Rao G A, Mahulikar S P 2002 Aeronaut. J. 106 629
Google Scholar
[2] Ball R E, Albrecht R S, Horne R L 2003 The Fundamentals of Aircraft Combat Survivability: Analysis and Design (2 nd ed.) (Reston, VA: AIAA) pp8–56
[3] Westwick P 2019 Stealth: The Secret Contest to Invent Invisible Aircraft (Oxford: Oxford University Press) pp5–42
[4] Singh H, Antony S, Jha R M 2016 Plasma-based radar cross section reduction (Singapore: Springer) pp1–46
[5] Knott E F 2012 Radar Cross Section Measurements (New York: Springer) pp12–36
[6] Knott E F, Schaeffer J R, Tuley M T 2004 Radar Cross Section (2 nd ed.) (Reston, VA: SciTech Publishing) pp4–22
[7] Kim S H, Lee S Y, Zhang Y, Park S J, Gu J 2023 Adv. Sci. 10 2303104
Google Scholar
[8] Ananth P B, Abhiram N, Krishna K H, Nisha M S 2021 Mater. Today Proc. 47 4872
Google Scholar
[9] Ye D, Wang Z, Xu K, Li H, Huangfu J, Wang Z, Ran L 2013 Phys. Rev. Lett 111 187402
Google Scholar
[10] Wang J, Yang R, Ma R, Tian J, Zhang W 2020 IEEE Access 8 105815
Google Scholar
[11] Liu Y, Zhao X 2014 IEEE Antennas Wireless Propag. Lett. 13 1473
Google Scholar
[12] Yu N, Genevet P, Kats Ma, Aieta F, Tetienne JP, Capasso F, Gaburro Z 2011 Science 334 333
Google Scholar
[13] Gao X, Han X, Cao W P, Li H O, Ma H F, Cui T J 2015 IEEE Trans. Antennas Propag 63 3522
Google Scholar
[14] Abdullah M, Koziel S 2022 IEEE Trans. Microw. Theory Techn. 70 264
Google Scholar
[15] Paquay M, Iriarte J C, Ederra I, Gonzalo R, Maagt P de 2007 IEEE Trans. Antennas Propag 55 3630
Google Scholar
[16] Chen W, Balanis C A, Birtcher C R 2015 IEEE Trans. Antennas Propag 63 2636
Google Scholar
[17] Sang D, Chen Q, Ding L, Guo M, Fu Y 2019 IEEE Trans. Antennas Propag 67 2604
Google Scholar
[18] Cui T J, Qi M Q, Wan X, Zhao J, Cheng Q 2014 Light Sci. Appl. 3 e218
Google Scholar
[19] Liu X, Gao J, Xu L, Cao X, Zhao Y, Li S 2016 IEEE Antennas Wireless Propag. Lett. 16 724
[20] Fu C, Han L, Liu C, Lu X, Sun Z 2021 IEEE Trans. Antennas Propag. 70 2352
[21] Li W, Huang N, Kang Y, Zou T, Ying Y, Yu J, Zheng J, Qiao L, Li J, Che S 2024 IEICE Electron. Express 21 20240246
Google Scholar
[22] Qi W j, Yu C, Du J l, Zhao Y j 2022 Int. J. RF Microw. Comput. Aided Eng. 32 23306
[23] Koziel S, Abdullah M 2020 IEEE Trans. Microw. Theory Techn. 69 2028
[24] Tao S, Pan X T, Li M K, Xu S H, Yang F 2020 IEEE J. Emerg. Sel. Top. Circuits Syst. 10 114
Google Scholar
[25] 杨欣雨 2023 硕士学位论文 (南京: 东南大学)
Yang X Y 2023 M.S. Thesis (Nanjing: Southeast University
[26] 袁方, 毛瑞棋, 高冕, 郑月军, 陈强, 付云起 2022 71 084102
Google Scholar
Yuan F, Mao R Q, Gao M, Chen Q, Fu Q Y 2022 Acta Phys. Sin. 71 084102
Google Scholar
[27] Yuan F, Wang G M, Xu H X, Cai T, Zou X J, Pang Z H 2017 EEE Antennas Wirel. Propag. Lett. 16 3188
Google Scholar
[28] Han X M, Xu H j, Chang Y P, Lin M, Zhang W Y, Xin W 2020 IEEE Access 8 162313
Google Scholar
[29] Zhou Y, Cao X Y, Gao J, Li S, Liu X 2017 IEEE Access 53 1381
[30] Katoch S, Chauhan SS, Kumar V 2021 Multimed. Tools Appl. 80 8091
Google Scholar
[31] Wang D S, Tan D P, Liu L 2018 Soft Comput. 22 387
Google Scholar
[32] Pan X, Xue L, Lu Y, Sun N 2019 Multimed. Tools Appl. 75 29921
[33] Dumoulin V, Visin F 2016 Comput. Sci. 03 07285
[34] Xu B, Wang N Y, Chen T Q, Li M 2015 arXiv: 1505.00853 [cs.LG]
[35] Bjorck N, Gomes C P, Selman B, Weinberger K Q 2018 arXiv: 1806.02375 [cs.LG]
[36] Srivastava N, Hinton G, Krizhevsky A, Sutskever I, Salakhutdinov R 2014 J. Mach. Learn. Res. 15 1929
[37] Dubey S R, Singh S K, Chaudhuri B B 2022 Neurocomputing 503 92
Google Scholar
[38] Zhou P, Xie X, Lin Z, Yan S 2024 IEEE Trans. Pattern Anal. Mach. Intell. 46 6486
Google Scholar
[39] Shi G, Zhang J, Li H, Wang C 2019 Neural Process. Lett. 50 57
Google Scholar
[40] Al-Kababji A, Bensaali F, Dakua S P 2022 arXiv: 2202.06373 [cs.CV]
[41] Yuan F, Xu H X, Jia X Q, Wang G M, Fu Y Q 2020 IEEE Trans. Antennas Propag. 68 2463
Google Scholar
DownLoad:
Metrics
- Abstract views: 551
- PDF Downloads: 25
- Cited By: 0
