Prediction of Metasurface Transmission Spectrum Based on An A-CNN-LSTM Approach

This paper employs an attention based convolutional neural network long short-term memory (A-CNN-LSTM) approach for the prediction of metasurface transmission spectrum. By inputting the pixel map of the metasurface structure, the trained A-CNN-LSTM model can quickly output corresponding transmission spectral data. The experimental results for different metasurface structures show that the A-CNN-LSTM has higher accuracy compared with the traditional CNN and CNN-LSTM models. And they agree well with traditional FDTD software simulation but can be much faster.