Neural Network‐Assisted Metasurface Design for Broadband Remote Invisibility

Abstract Traditional metasurface‐based invisibility techniques typically involve attaching meticulously designed structures directly onto the surface of the object to be concealed, making the approach highly dependent on the object's geometry and severely limiting its mobility and environmental interaction. In this work, a broadband invisibility device is proposed designed using deep learning and differential optimization, enabling the concealment of objects located at a certain distance from the metasurface. The device achieves remote invisibility across both the X‐ and Ku‐band simultaneously. A deep learning model is employed to efficiently simulate the electromagnetic response of unit cells, while a differential optimization algorithm tailors the design for various tasks to obtain optimal unit configurations. To validate the effectiveness of this approach, objects of varying sizes are randomly placed in front of the device, and the system consistently demonstrated excellent invisibility performance across the 8–18 GHz frequency range. The study showcases the precise control of broadband electromagnetic waves by metasurfaces and offers new insights for the development of other intelligent optical devices, such as smart concentrators and optical camouflage systems.