Design of bistatic radar cross section reduction metasurface based on convolutional neural networks

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 development of radar detection technologies, RCS reduction has become a forefront research topic in radar stealth, aiming to minimize target detectability. With the maturity of radar networking technology, the bistatic radar RCS reduction is becoming increasingly important in future electromagnetic stealth countermeasures compared with the monostatic radar RCS reduction. Artificial electromagnetic metasurfaces have introduced innovative technical approaches for realizing the bistatic radar RCS reduction. However, current metasurface designs still face challenges related to inefficiency and suboptimal performance, mainly 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. This approach integrates convolutional feature extraction, residual enhancement, and fully connected optimization modules with 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 can achieve over 10 dB of Bistatic Radar RCS reduction in a frequency range from 7.26 GHz to 10.74 GHz. The method also ensures uniform diffuse reflection across the full space for various incidence angles (30 degrees, 45 degrees, 60 degrees). Compared with traditional optimization algorithms, this method enhances RCS reduction by 17.2% while significantly improving computational efficiency. This approach provides a promising new technical paradigm for achieving full-space electromagnetic stealth in advanced weapon systems.