On the Exploitation of Deep Reinforcement Learning for Microwave Inverse Scattering

This work explores the use of Deep Reinforcement Learning (DRL) for solving electromagnetic (EM) inverse scattering problems (ISPs). Specifically, a pixel-wise DRL model is trained to reconstruct the contrast distribution of an unknown object within an inaccessible domain by iteratively updating its estimates based on interactions with the surrounding environment, guided by actions and rewards derived from scattered field data. The results demonstrate the model capability to learn an effective inversion strategy, highlighting its potential for integration into highly-relevant microwave imaging (MI) applications, such as biomedical imaging.