Flexible conformal polarization-sensitive structural color optical switch

Traditional metasurface structural color design faces limitations in computational efficiency and optical switching mechanisms. This study introduces a bidirectional deep neural network (Bi-DNN) platform for the inverse design of a flexible polarization-sensitive color switch. The trained model enables simultaneous bidirectional prediction of structural parameters and output colors, achieving prediction speeds of 64 ms per sample-three orders of magnitude faster than conventional finite-difference time-domain (FDTD) simulations. Under mechanical tuning (-10% to 50% strain), the metasurface achieves a color gamut covering 92% sRGB while maintaining target color accuracy (Δ<i>E</i> < 2) via Bi-DNN driving. The device enables color display or hiding through orthogonal linear polarization switching, achieving a maximum extinction ratio of 17.1 dB within the visible spectrum. Angularly stable performance preserves color fidelity over incidence angles of 0°-20°. This work establishes a new paradigm for high-performance optical switches and advances inverse-design foundations for directional color metasurfaces, with applications in next-generation displays, smart wearables, and optical encryption systems.