光栅反散射问题的神经网络方法

Neural Network Method for Inverse Diffraction Grating Problem

针对测量的近场数据来研究光栅形状重构问题,提出了一种基于端对端结构的神经网络方法。该方法是一种循环神经网络,采用序列对序列的方式进行计算。网络模型以近场数据作为输入,以光栅形状参数作为输出,先利用编码端对输入的近场数据进行特征提取,再通过Adam算法更新模型权重,最后使用解码端进行光栅形状参数的反演。此外,模型利用多个门控循环单元从近场数据中提取近场特征,并将该特征引入到解码端中,为反演光栅形状参数提供了更多的特征参考,进一步提高反演效果。数值实验说明该方法可以有效地重构光栅的形状。

In this paper,we study the problem of grating shape reconstruction based on the measured near-field data,and propose a neural network method based on the end-to-end structure.This method is a cyclic neural network,which uses the sequence to sequence method to calculate.The network model takes the near-field data as the input,and the shape parameters of the grating as the output.Firstly,the encoder is used to extract the features of the input near-field data,and then the Adam algorithm is used to update the model weight.Finally,the decoder is used to invert the shape parameters of grating.Moreover,the model uses multiple gating cycle units to extract near-field features from the near-field data and introduces this feature into the decoding end,which provides more feature reference for the inversion grating shape parameters and further improves the inversion effect.Numerical experiments show that this method can reconstruct the shape of grating effectively.