Single-Sensor Source Localization Using Electromagnetic Time Reversal and Deep Transfer Learning: Application to...

Area of Science:

• Electromagnetics
• Signal Processing
• Artificial Intelligence

Background:

• Electromagnetic Time Reversal (EMTR) is established for source localization.
• Existing methods often require multiple sensors, limiting efficiency and increasing complexity.

Purpose of the Study:

• To develop a novel, single-sensor technique for electromagnetic source localization.
• To integrate Electromagnetic Time Reversal (EMTR) with Machine Learning (ML) for enhanced accuracy and reduced sensor requirements.

Main Methods:

• Utilized the 2D-Finite-Difference Time-Domain (2D-FDTD) method to generate electromagnetic field profiles as images.
• Employed transfer learning with a pre-trained VGG-19 Convolutional Neural Network (CNN) for feature extraction from simulation-generated images.
• Applied the combined EMTR-ML approach to localize radio frequency (RF) sources and lightning impulses.

Main Results:

• Successfully demonstrated accurate 2D localization of electromagnetic sources using a single sensor.
• Achieved precise 2D lightning localization in the Säntis region using experimental data, outperforming traditional multi-sensor systems.
• Validated the efficacy of using pre-trained CNNs with simulation-generated data for the first time.

Conclusions:

• The novel EMTR-ML methodology significantly reduces the number of sensors required for electromagnetic source localization.
• This approach offers a more efficient and accurate solution for locating diverse electromagnetic sources, including lightning.
• The study highlights the potential of applying deep learning techniques to simulation-based electromagnetic analysis.