Enhancing DORT Method Performance in Time-Reversal Microwave Imaging Through Denoising Autoencoder

Abstract

We investigate the impact of noise on time-reversal imaging and propose an approach that significantly enhances the detection of objects in noisy environments. Our method involves the decomposition of the time-reversal operator at a single frequency, known for its sensitivity to noise. We utilize a specific autoencoder architecture to denoise the generated dataset from a multi-static data matrix (MDM), effectively separating the signal sub-space from the noise sub-space, even at low signal-to-noise ratios (SNRs) ranging from −5 dB to high levels of SNR. This dataset is generated by simulating scatterers mounted at various locations within a two-dimensional (2D) grid, each with different SNRs.