DDA-based Proxy Model for 3D Snow Scattering using Neural Networks

This paper studies a neural network proxy model for the three-dimensional snow scattering problem. Scattering coefficients can be used for large-scale scene-level electromagnetic simulation. To create a dataset for the neural network, the discrete dipole approximation (DDA) algorithm for the half-space Green's function was used to solve the full wave scattering of three-dimensional snow. A fully connected neural network was trained on the dataset to obtain a proxy model that can quickly predict scattering results. Simulation experiments were conducted between the neural network proxy model and the DDA algorithm. The experimental results show that the neural network can significantly reduce simulation time while maintaining high accuracy.