基于通道注意力增强网络的高效消色差超透镜逆向设计OA

Rapid and accurate prediction of the phase spectral response of meta-units is essential for the in-verse design of metalenses.Conventional electromagnetic simulation methods are computationally inten-sive and inefficient,making them inadequate for the intelligent design of metasurface devices.To address this limitation,a convolutional neural network incorporating an efficient channel attention mechanism(EC-ANet)is proposed for forward prediction of the phase spectrum of pixelated meta-atom structures.Train-ing and validation on a simulated dataset demonstrate that the mean absolute error of phase prediction with-in the design bandwidth is below 0.06 rad.The trained ECANet is further integrated into a particle swarm optimization(PSO)algorithm to establish an efficient inverse design framework for achromatic metalens-es.Simulation results indicate that the designed metalens achieves a focal length variation of less than 2%across the operating bandwidth,demonstrating excellent achromatic focusing performance.By replacing conventional finite-difference time-domain(FDTD)simulations with rapid phase predictions from EC-ANet,the proposed framework improves computational efficiency by approximately two orders of magni-tude while maintaining design accuracy.This work presents a novel and efficient approach for the intelli-gent design of pixelated metasurface devices in the visible regime,with significant potential for applications in integrated optics,micro-imaging systems,and related fields.