Numerical investigation of a CNN-based approach to reconstruct turbulent electric fields in laser-driven plasmas

Diagnosing turbulent electromagnetic fields in laser-driven plasmas is challenging because ion radiography cannot easily resolve the electric and magnetic field components. To address this limitation, we propose and numerically investigate a reconstruction method based on encoder–decoder convolutional neural networks. We construct a simulation dataset for training and testing the neural networks by pairing electric fields generated in hydrodynamic simulations with corresponding synthetic ion radiographs obtained from test particle simulations. Using the resulting dataset, we systematically investigate the influence of several key factors on the reconstruction, including the addition of synthetic noise, variations in the probing ion flux, and the use of different input data types.