Metasurface design method based on deep learning and hologram weight partition regulation and control

• Main Authors: GU JIANQIANG, GUO YUNXIN
• Format: Patent
• Language: Chinese; English
• Published: 26.04.2024
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; IMAGE DATA PROCESSING OR GENERATION, IN GENERAL; INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTEDFOR SPECIFIC APPLICATION FIELDS; PHYSICS

The invention discloses a super-surface design method based on deep learning and hologram weight partition regulation and control, which comprises the following steps of: calculating electromagnetic response frequency spectrums of super atoms under different structure parameters by means of an electromagnetic numerical simulation tool to form a data set; based on a deep learning framework, training a forward network from structure parameters to corresponding electromagnetic response spectrums; the gradient of the forward network is frozen, and a reverse network formed by a convolutional network is connected in series for training, so that when target images at different positions are input into the reverse network, the network automatically assigns structural parameters of the metasurface; inputting the structure parameters of the metasurface into a forward network to predict two-dimensional amplitude distribution, and comparing the two-dimensional amplitude distribution with a target map; and when an analyti