Using Tandem Neural Networks to Observe Jones Matrix Parameters
Cong Liu, Yueyi Yuan, Yuxiang Wang, Kuang Zhang · 2024
Deep learning, with its powerful computing and processing capabilities for data, is increasingly being applied in the design and regulation of metasurface and other tasks. This article proposes a method for freely adjusting the eight degrees of freedom of the Jones matrix through deep learning. Using tandem neural networks (TNN), reverse design of metasurface units for different Jones matrices to achieve polarization control. The proposed method can greatly accelerate the implementation of polarization control and metasurface design.