Machine Learning Exercises on 1-D Electromagnetic Inversion

This work aims to enhance our fundamental understanding of how the measurement setup that is used generate training and testing data sets affects accuracy machine learning algorithms attempt solve electromagnetic inversion problems solely from data. A systematic study carried out on a 1-D semi-inverse problem, which estimating electrical permittivity values planarly layered medium with fixed layer thicknesses assuming different receiver–transmitter antenna combinations in terms location numbers. The solutions obtained four methods, including neural networks, compared physics-based solver deploying Nelder–Mead simplex method achieve iteratively. Numerical results show that: 1) deep-learning outperforms other techniques implemented this study; 2) increasing number antennas placing them as close possible domain interest increase accuracy; 3) for created random grids lead more efficient than uniform grids; 4) multifrequency few can accurate single-frequency setups several antennas.