Inverse design of a broadband low-RCS coding metasurface via synergistic DNN and NSGA-II optimization

Metasurface-mediated radar-cross-section suppression underpins next-generation stealth. Traditional trial-and-error synthesis is slow and narrowband. Here, a deep neural network is fused with NSGA-II to enable inverse design of broadband, low-RCS coding metasurfaces without repeated full-wave solvers. The network swiftly maps 4-bit coding sequences to far-field scattering, while the genetic optimizer globally balances RCS minimization with spectral breadth. The engineered random-coding array delivers a 37.5 dB peak and 15.7 dB mean RCS drop under normal linear polarization across the 10–20 GHz X–Ku band, sustaining reductions exceeding 10 dB between 13.5 and 20 GHz. This automated photonic framework accelerates practical stealth deployment.