Physics-Informed Supervised Residual Learning for 2-D Inverse Scattering Problems