Two-Step Contrast Source Learning Method for Electromagnetic Inverse Scattering Problems

Area of Science:

• Electromagnetics
• Computational Physics
• Applied Mathematics

Background:

• Electromagnetic inverse scattering problems (EM-ISPs) are crucial for material characterization but are often ill-posed and nonlinear.
• Traditional methods face challenges in achieving both high accuracy and real-time reconstruction for EM-ISPs.
• Existing techniques struggle with high-contrast objects, limiting practical applications.

Purpose of the Study:

• To develop an accurate and efficient method for solving 2D full-wave electromagnetic inverse scattering problems.
• To address the limitations of traditional iterative and non-iterative methods in terms of speed and accuracy.
• To enable real-time quantitative microwave imaging, particularly for high-contrast objects.

Main Methods:

• A two-step contrast source learning approach integrating convolutional neural networks (CNNs) is proposed.
• The first step utilizes a CNN-based contrast source network for initial contrast estimation.
• The second step employs a U-Net architecture for refining the initial contrast estimate.

Main Results:

• The proposed method significantly improves accuracy and robustness for EM-ISPs, including high-contrast scenarios.
• Near real-time imaging capabilities are achieved, overcoming the speed limitations of conventional techniques.
• Numerical simulations validate the effectiveness of the two-step learning approach.

Conclusions:

• The two-step contrast source learning approach offers a powerful solution for full-wave EM-ISPs.
• This method effectively combines traditional and deep learning techniques for enhanced performance.
• The approach shows significant promise for advancing real-time quantitative microwave imaging of complex objects.