Deep-learning-assisted inverse design of polarization-multiplexed structural color filters with ultrahigh saturation based on all-dielectric metasurface

Abstract

Although great progress has been made for metasurface-based structural color filters, there are still significant challenges in encoding metasurfaces with complex color images, especially in high saturation or multi-channel cases, where traditional iterative optimization calculations demand substantial computational resources. Here, we proposed a dual-channel multilayered all-dielectric metasurface that can generate structural colors accounting for 224% Standard RGB space, 168% Adobe RGB space and 78% of the 1931 CIE chromaticity space under vertically-polarized incidence. A deep learning-based designing approach is introduced to significantly reduce the computing consumption while maintaining high accuracy. Numerical simulations further demonstrated that the proposed design method is effective for encoding both complementary and independent images with high saturation and wide color gamut. Therefore, the proposed color filters may find potential applications in image display, information encryption and optical anti-counterfeiting.

Keywords

• Metasurface
• Structural color
• Deep learning
• Polarization