Optically Transparent Metasurface Absorber Optimized by LSTM Neural Network for Broadband Ultra-low Microwave Scattering

In this paper, an intelligent design optimization method based on Long Short Term Memory (LSTM) neural networks is introduced, aiming at designing discrete pixelated meta-atom structure with high efficient microwave absorption across a wide band. By rapid generation and selection of metaatom pairs with both high absorption and stable broadband phase differences, an absorption scattering integrated coding metasurface realized with optically transparent ITO film can be obtained by random arranging the two kinds of meta-atoms. Simulation and experiment have demonstrated the capability of achieving a 20 dB RCS reduction across an octave band from 10.5 to 27 GHz. The proposed machine-learning based metasurface design could boast advantages including high efficient absorption, broad bandwidth, light weight, optical transparency, and low profile, holding promise for advancing the development of low radar detectable technique.