The Physics-Informed Neural Network (PINN) corresponds to a machine learning strategy to approximate the solution of partial differential equations by in- cluding the residual PDE in the loss function. In a previous work, we found that adding physical coefficients as predictor variables in a PINN for boundary layer linear problems improves the accuracy of the approximation when com- paring it with the approximate solution obtained from PINNs that use only spatio-temporal inputs. This work explores this same strategy for the time-dependent Maxwell linear equations in case electric and magnetic fields present highly oscillatory behavior. Extensive numerical experiments assess this strategy by simulating electromagnetic fields in heterogeneous media with high permittivity variability in small spatiotemporal regions of the domain