Bayesian-neural-network accelerated design of multispectral-compatible camouflage layer with wide-band microwave absorption, customized infrared emission and visible transparency

Metasurfaces with customizable multi-spectrum compatibility have attracted a great deal of attention due to the increasing applications of stealth and camouflage. Compared with single-spectrum and fixed-scene camouflage metasurfaces, a camouflage metasurface with the designability and spectral-independence can adapt to more complex environments. However, this drastically increases the design complexity and time cost of metasurfaces. Here, we propose a method to design multispectral compatible function layer of camouflage which can achieve customizable wideband microwave absorption, visible transparency and selective infrared emission, simultaneously. The design of the camouflage layer is optimized and accelerated by Bayesian-neural network, which can be implemented based on a very limited amount of prior data and can greatly reduce the design complexity and time. To verify our method, an optical-transparent metasurface with digital infrared camouflage and low backward scattering was designed and fabricated. The simulation and experimental results well demonstrate the performance of multispectral compatible camouflage of the metasurface. We further explore the angular stability of the camouflage layer to the incident angle varying from 0 to 60° at microwaves. Our proposed design scheme provides a novel method to design spectrum-individual and devisable metasurfaces, which can be applied to camouflage in complex spectrum background.