Supervised Descent Method for Full-wave Microwave Imaging

Abstract

This work studies the application of the supervised descent method (SDM) to two dimensional full-wave microwave imaging. In SDM-based inversion, the reconstructed models are updated based on the learned descent directions of the cost function and the residual between simulated and measured data. It provides a flexible method to incorporate prior information into the inversion that can improve both the quality of reconstruction and the efficiency of reconstruction. In this work, we studied both pixel- and model-based SDM inversion. An online restart scheme is applied to further reduce data misfit. The inversion algorithm is validated using both synthetic and experimental data. The results show good generalization ability of this machine-learning-based inversion scheme, and the reconstruction can be achieved in both high efficiency and accuracy.