High-Resolution Programmable Metasurface Imager Based on Multilayer Perceptron Network

摘要

In the data-driven society, fidelity and accuracy of automatic decisions behindthe scene rely fundamentally on a solid data or imaging acquisition system.However, conventional microwave imagers are inadequate relating to theirresolution and noise capability, mainly due to the limited aperture size andrigid working principle. Here, a programmable metasurface imager with highresolutionand anti-interference performance is proposed. By implementingthe structure of multilayer perceptron network in the analog domain, themetasurface-based microwave imager intelligently adapts to different datasetsthrough illuminating a set of designed scattering patterns that mimic thefeature patterns. A prototype imager system working at microwave frequencyis designed and fabricated. The accuracy rate rises by 18 under the classificationtask of MNIST dataset, with a decline in the reconstruction imagingerror. The authors experimentally demonstrate that the resolution to distinguishstrip patterns goes beyond to one-fifth of the equivalent wavelength onthe target plane.