Machine Learning-Assisted Design of Infrared-Radar-Visible Light Compatible Stealth Flexible Metamaterial Structures.

Multispectral compatible stealth is crucial for modern detection environments. Despite conflicts among stealth mechanisms, current strategies lack effective compatibility. This study introduces a flexible gradient metamaterial structure inspired by bionic design and machine learning's forward prediction mechanism. This structure achieves spectral decoupling and enhances stealth performance across infrared, radar, and visible light spectra. The multispectral compatible stealth metamaterials (MCSM) comprise a radar infrared compatible stealth layer (RICSL) with high transmittance (78.56%) and a pixelated tunable visible light camouflage pixel layer (VLCPL). By adjusting the hexagonal patches' filling rate on the infrared stealth layer (ISL), a blend of low infrared emissivity (0.2) and high microwave transmission efficiency is achieved. The structure ensures efficient microwave absorption through gradient impedance transition and multiscale loss mechanisms, with absorption efficiency exceeding 90% in the measured wideband range of 6.34 to 24.91 GHz, along with polarization insensitivity and angular stability. The VLCPL can adapt patterns to mimic jungle, ocean, and desert environments. In practical settings, these metamaterial structures demonstrate flexible adaptive features in infrared and visible light stealth effects, paving the way for innovative multispectral compatible stealth technologies.