Parameter Reconstruction of 3D Maxwell’s Equations Based on Point Cloud

A three-dimensional deep learning reconstruction algorithm based on point cloud is proposed to reconstruct the parameters of the three-dimensional Maxwell’s equations, which is embodied in the reconstruction of the permittivity of the three-dimensional anisotropic Maxwell’s equations. The point cloud data used in model, can carry more physical field information than voxel data, with simpler storage. By sharing the feature extraction layer and the dual-task output branch, the geometric structure and permittivity can be reconstructed simultaneously. It performs well in electric field reconstruction by combining three-dimensional convolution network, custom loss function and flexible training mechanism. The deep learning network architecture and custom training cycle are used in the training, with a variety of optimization strategies to ensure efficient and stable. Numerical experiments show that the dual reconstruction of electric field geometry and dielectric parameters can be realized by using different numbers of point cloud data, and high accuracy can be achieved.