Convolutional Neural Network-Based Electromagnetic Imaging of Uniaxial Objects in a Half-Space

Area of Science:

• Electromagnetic imaging
• Artificial intelligence applications
• Geophysical exploration

Background:

• Imaging buried objects in half-space environments is challenging due to limited measurement angles.
• Uniaxial objects require specialized electromagnetic imaging techniques.
• Existing methods face difficulties with data acquisition and reconstruction accuracy.

Purpose of the Study:

• To apply artificial intelligence (AI) for improved electromagnetic imaging of uniaxial objects in half-space.
• To compare the performance of the dominant current scheme (DCS) and backpropagation scheme (BPS) for permittivity reconstruction.
• To evaluate the effectiveness of Convolutional Neural Networks (CNNs) in this imaging context.

Main Methods:

• Simultaneous emission of Transverse Magnetic (TM) and Transverse Electric (TE) waves for object illumination.
• Computation of initial permittivity distribution using DCS and BPS.
• Training Convolutional Neural Networks (CNNs) with the computed permittivity data.

Main Results:

• DCS demonstrated superior generalization capabilities compared to BPS under identical conditions.
• The DCS method showed enhanced noise immunity in reconstructing object permittivity.
• Both DCS and BPS, when integrated with CNNs, proved effective for dielectric constant distribution reconstruction.

Conclusions:

• AI, particularly CNNs trained with DCS, offers a robust solution for electromagnetic imaging of buried uniaxial objects.
• The DCS method is recommended for its improved accuracy and resilience to noise in half-space imaging scenarios.
• This study validates the potential of advanced computational schemes for subsurface object characterization.