Electromagetic time reversal (EMTR) and Machie Learning (ML)

Implementation of the hybrid EMTR/ML methods for a range of applications including electromagnetic source localization, source recunstruction, and microwave imaging.

1. Single sensor EM Source localization using EMTR and deep transfer learning

Please refer to this paper for detailed explanation of the method.

Getting started

Clone this repository:

git clone [email protected]:fanfeum/EMTR_ML.git
cd EMTR_ML

Loadig the dataset, extracting the feature vectors, and training the model

• Open jupyter notebook in the current directory:

jupyer notebook

• Open and run '1sesnor.ipynb':

  1. Specify the path to the source images and how many images to use for training and evaluation.
  2. Specify which coordinante of the source to estimate (whether x- or y- coordinante).
  3. Re-run the code for the other source coordiante.
  4. The model's estimation for the source coordiantes along with the ground truth source positions will be automatically saved for further inference.

Visualizing the evaluation results

Run 'figureMaker.m' in Matlab:

• Using the saved model's outputs and the target positions, this code will generate the followig plots:
  1. Estimation of the x- and y-coordinate of the randomly selected source locations.
  2. Histogram of the location error
  3. Average location error presented as a heatmap chart inside the detection region