A Triple-Network Dual-Driven Framework for Plasma Electron Density Inversion

We propose a triple-network, dual-driven deep learning framework for plasma electron density inversion. It includes the Antenna Decoupling Network (ADNet), which maps antenna reflection coefficients to plane-wave equivalents; the Initial Estimation Network (IENet), which provides coarse electron density predictions; and the Data-Enhanced Physics-Informed Neural Network (DE-PINN), which performs inversion with Maxwell-based physical constraints. By combining data-driven initialization and physics-informed modeling, the framework enhances both accuracy and generalization. Numerical results validate its effectiveness in reconstructing electron density profiles for non-uniform plasmas.