A Convolutional Neural Network with Multifrequency and Structural Similarity Loss Functions for Electromagnetic...

Area of Science:

• Electromagnetic imaging
• Artificial intelligence applications
• Anisotropic object characterization

Background:

• Electromagnetic imaging is crucial for detecting subsurface objects using magnetic anomaly sensing.
• Current methods face challenges in accurately characterizing complex anisotropic materials.

Purpose of the Study:

• To apply artificial intelligence (AI) to enhance electromagnetic imaging of anisotropic objects.
• To improve the accuracy and stability of subsurface object reconstruction.

Main Methods:

• Utilized multifrequency scattered fields and the backpropagation scheme (BPS) for initial dielectric constant calculation.
• Employed a convolutional neural network (CNN) with adaptive moment estimation (ADAM) for refined image reconstruction.
• Introduced an improved loss function combining structural similarity index measure (SSIM) and root mean square error (RMSE).

Main Results:

• The enhanced CNN with the improved loss function significantly improved image quality.
• Simulations considered transverse electric (TE) and transverse magnetic (TM) wave noise interference.
• Multifrequency reconstructions demonstrated superior stability and precision compared to single-frequency methods.

Conclusions:

• AI-driven electromagnetic imaging offers a powerful tool for characterizing anisotropic objects.
• The proposed CNN approach with an optimized loss function enhances reconstruction fidelity.
• Multifrequency analysis is key to achieving robust and precise subsurface imaging.