Quantitative Microwave Imaging Using Deep Learning Network Guided by Plane Wave Equation

Accurately characterizing material properties, particularly the spatial distribution of permittivity, is crucial across diverse domains such as medical imaging, nondestructive testing, and materials science. This work introduces an innovative strategy to tackle the inverse problem of deducing the permittivity distribution within a medium by leveraging time-dependent data of the electric field distribution. The approach utilizes a neural network trained with guidance from the wave equation, embedding the fundamental physics of wave propagation within the network architecture. This integration empowers the network to assimilate domain-specific knowledge during training, combining deep learning capabilities with physics-based constraints. This hybrid framework establishes a robust relationship between the time changing electric field distribution and the underlying permittivity distribution, effectively solving the complex inverse problem. By training on a comprehensive dataset, the neural network discerns intricate variations in spatial permittivity from the intricate temporal evolution of the electric field. Results validate the effectiveness of this approach, showcasing impressive accuracy in the reconstruction of the permittivity distribution.