Inverse design of polarization-independent structural color based on a coding metasurface via a bidirectional artificial neural network

Abstract

A polarization-independent structural color based on a coding metasurface is reverse designed via a bidirectional neural network. A forward prediction network from coding metasurface structures to colors is constructed by introducing a bidirectional long short-term memory (Bi-LSTM) model. Based on this model, a bidirectional neural network training method is adopted to achieve reverse design from the target color to the optimal structure of the coding metasurface. The results show that the method can achieve 91% accuracy for forward prediction of color and 92% accuracy for inverse design of the structure. In addition, the coding metasurface structure has quadruple rotational symmetry, which realizes that the structural color is independent of the polarization of the incident light. This study provides a novel polarization-independent structural color design scheme, providing a new path for the application and development of structural colors.