江苏大学学报（自然科学版）2025，Vol.46Issue(4)：438-443,6.DOI:10.3969/j.issn.1671-7775.2025.04.010
基于卷积神经网络的窄线宽光谱结构参数优化
Optimization of narrow linewidth spectral structure parameters based on convolutional neural networks
摘要
Abstract
To solve the problem of excessive time spending on modeling,calculation and simulation optimization in conventional optical structure design,the novel optimization method combining convolutional neural networks and genetic algorithms for narrow linewidth spectral structure parameters was proposed.Using the Y-shaped all-dielectric metasurface structure as model,4 096 datasets were generated by finite difference time domain(FDTD)simulation to train the forward prediction network.The trained network was further combined with the genetic algorithm to optimize the parameters of metasurface structure.The simulation results show that the loss value of the trained prediction network on the test set is only 5.6×10-4.Compared to the original dataset,the minimum full width at half maximum(FWHM)obtained by combining the optimization algorithm is reduced by 0.040 nm.Compared with the traditional methods,the proposed method can enhance the optimization efficiency and effectiveness of complex metasurface structures.关键词
超表面/微纳结构设计/Fano共振/深度学习/卷积神经网络/优化算法Key words
metasurface/micro-nano structure design/Fano resonance/deep learning/convolutional neural networks/optimization algorithm分类
信息技术与安全科学引用本文复制引用
富小鸥,王原丽,杜庆国,付琴..基于卷积神经网络的窄线宽光谱结构参数优化[J].江苏大学学报（自然科学版）,2025,46(4):438-443,6.基金项目
国家自然科学基金资助项目(62075173) （62075173）
