Low-Frequency Electromagnetic Characterization of Layered Media Using Deep Neural Network

The problem of determining the dielectric properties and thickness of materials using electromagnetic waves can be found in different geophysical applications. In such scenarios, a radiating source is placed on the top interface, and a receiver is moved along a preferred direction to measure the electromagnetic fields. In this work, we have considered such a scenario with low-frequency electromagnetic waves. We have trained a deep neural network to learn the complex relationship between the field values and substrate thickness and dielectric constants. Then we have used that network to estimate these quantities from field value measurements. The results show very low mean absolute percentage errors for both noise-free and noisy training data as long as the noise level of the test data is not too high. A successful realization of the proposed method can help monitor the characteristics of geophysical substances like glacier, sea ice etc.