Microwave Imaging of Anisotropic Objects by Artificial Intelligence Technology

Area of Science:

• Electromagnetics
• Artificial Intelligence
• Microwave Imaging

Background:

• Biaxial anisotropic scatterers present complex challenges for microwave imaging due to differing dielectric constants.
• Transverse electronic (TE) polarization waves encounter higher nonlinearity than transverse magnetic (TM) waves, complicating image reconstruction.
• Existing methods struggle to accurately reconstruct microwave images from scattered field information of anisotropic objects.

Purpose of the Study:

• To develop and validate an artificial intelligence-based approach for microwave imaging of biaxial anisotropic objects.
• To address the complexities associated with TE polarization in microwave imaging.
• To compare the efficacy of different initial image reconstruction schemes when used with a CNN.

Main Methods:

• Utilized the dominant current scheme (DCS) and back-propagation scheme (BPS) for initial image estimation.
• Applied a trained convolution neural network (CNN) to regenerate and refine microwave images.
• Conducted numerical simulations to evaluate the CNN's performance and generalization ability.

Main Results:

• The CNN demonstrated good generalization capabilities, even with limited training data.
• The proposed CNN-based method successfully reconstructed microwave images of biaxial anisotropic objects.
• Comparison showed the dominant current scheme (DCS) outperformed the back-propagation scheme (BPS) when integrated with the CNN.

Conclusions:

• Convolution neural networks offer a promising solution for complex microwave imaging problems involving anisotropic objects.
• The AI-driven approach effectively overcomes the high nonlinearity challenges of TE polarization.
• The dominant current scheme (DCS) is a more suitable precursor for CNN-based microwave image reconstruction of anisotropic scatterers compared to BPS.