Research on Electromagnetic Scattering and Inverse Scattering of Target Based on Transfer Learning Physics-Informed Neural Networks

Traditional electromagnetism is significantly more computationally complex and challenging when faced with, for example, more complex high-dimensional problems. In recent years, efficient data-driven methods such as deep learning have become powerful instruments for studying electromagnetic scattering and the inverse scattering of targets. Physical-informed Neural Networks (PINNs) can embed model equations as a physical driving force into the neural network, and are able to learn more generalizable models with small datasets compared to traditional data-driven neural networks. At the same time, transfer learning can accelerate the training speed of neural networks as well as their generalization ability. This paper combines transfer learning with PINNs and proposes Transfer Learning Physics-Informed Neural Networks (TL-PINNs), which greatly improves the training speed and generalization ability of PINNs and has broad application prospects.