A multiscale residual U-net architecture for super-resolution ultrasonic phased array imaging from full matrix capture data.

This article focuses on the development and evaluation of FMC-Net, a deep learning (DL) architecture designed to reconstruct high-resolution ultrasonic phased array (UPA) images directly from full-matrix capture (FMC) data. FMC-Net employs an encoder-decoder structure with multiscale residual modules and residual skip connections to model nonlinear wave-matter interactions, enabling improved imaging resolution beyond the Rayleigh diffraction limit and enhanced visualization of subwavelength defects. Trained on simulated and experimental datasets, FMC-Net demonstrated superior performance compared to conventional beamforming methods such as the total focusing method (TFM) and wavenumber algorithm, achieving faster image reconstruction and better defect characterization in nondestructive evaluation (NDE) applications. The approach shows promise for real-time deployment in various ultrasonic array imaging contexts by shifting computational complexity to the training phase while enabling efficient inference.