High Resolution Microwave Tomographic Imaging Using Low Frequency Data and Artificial Neural Networks

The spatial resolution obtained from microwave tomographic imaging (MTI) is directly related to the frequency of the electromagnetic field used to illuminate the imaging domain. Although higher frequencies have the potential to yield higher spatial resolution, high frequency waves tend to attenuate more, especially in lossy media, such as, biological tissue. By contrast, lower frequency electromagnetic waves tend to penetrate the imaging domain more, yet at the cost of lower spatial resolution. In this work, we propose an ingenious technique to estimate higher frequency data using lower frequency measurements for MTI. Specifically, higher frequency data are estimated using a multi-layer perceptron (MLP) that uses the lower frequency data as input. For improved performance, the imaginary part (loss) of the bulk medium is set to a lower value than the actual sample for the higher frequency while generating the training data for the MLP. This approach dramatically increases the frequency that can be employed to reconstruct the image. The proposed technique is validated in 2D imaging scenarios with anomalies present in a homogeneous medium. In particular, the formulation of line source scattering from dielectric circular cylinders is employed to generate the dataset to train and test the MLP. It is demonstrated that the method yields high-resolution images using low-frequency data for these scenarios.