A model-based regularized inverse method for ultrasonic B-scan image reconstruction

Abstract

Ultrasonic B-scan imaging is affected by the acoustic diffraction and electrical effects in nondestructive testing (NDT), resulting in insufficient lateral and temporal resolution for defect characterization. The minimum mean squared error (MMSE) method can improve the resolution by inversing a linear imaging model, which takes the acoustic diffraction and electrical effects into account, and achieve higher resolution than the synthetic aperture focusing technique (SAFT). However, its computation efficiency and resolution improvement are unsatisfactory due to the hypothetical Gaussian distribution of defects. To overcome these problems, a model-based regularized inverse method for ultrasonic B-scan image reconstruction is proposed. Benefitting from the sparse distribution of defects in NDT applications, the proposed method formulates an inverse objective function composed of -norm as well as -norm, and the sparse reconstruction by a separable approximation (SpaRSA) algorithm is adopted to obtain the optimal solution. The performance of the proposed method is evaluated by B-scan imaging of two 0.3 mm steel wires conducted both in simulation and experiment. The results verify that the proposed method improves the lateral and temporal resolution simultaneously with high computation efficiency.