Reconstructing Electromagnetic Parameters Based on Convolutional Neural Network for the Groove Scattering with Impedance Boundary Condition

Abstract

The reconstruction of electromagnetic parameters of the groove scattering is widely applied in the military and engineering fields. However, the inherent nonlinearity and ill-posed nature of the problem bring a big challenge to reconstruction. To over the difficulty, we develop a real-valued convolutional neural network (RV-CNN) and a complex-valued convolutional neural network (CV-CNN) to reconstruct the electromagnetic parameters of the open groove with impedance boundary. First the scattered field data of the open groove is obtained using the Petrov-Galerkin finite element interface method. Then, the RV-CNN separately extracts the magnitude and phase of the scattered field data, and introduces them as independent input channels into the network architecture, with the ELU activation function selected to align with the char acteristics of the phase data. In contrast, the CV-CNN directly takes the complex-valued scattered field data as input channels and performs optimization derivation based on a complex-valued loss function. This approach not only accelerates the convergence speed of the model but also enhances the overall reconstruction capability. Numerical experimental results demonstrate that both RV-CNN and CV-CNN can achieve accurate reconstruction results with a smaller volume of training data and reflect good general ization performance for the reconstruction of electromagnetic parameters of the open groove filled with homogeneous and inhomogeneous media under impedance boundary.