Imagerie ultrasonore par réseaux antagonistes génératifs conditionnels (cGAN)

Abstract

The Full Matrix Capture (FMC) combined with the Total Focusing Method (TFM) are often considered as the gold standard in ultrasonic nondestructive evaluation. However, this method is not always convenient because of the amount of data required and the time required to gather the data. Indeed, when it comes to high cadence inspections, gathering the FMC and processing it may take too long. This study proposes to replace conventional FMC acquisition and TFM processing with a single zero-degree Plane Wave (PW) insonification and a conditional Generative Adversarial Networks (cGAN) trained to produce TFM-like images. Three models with different architecture and loss formulations were tested in different scenarios and their performances were compared with TFM images. The proposed cGAN were able to recreate the important features of the images and also improve the contrast in more than half the reconstructions in comparison with conventional TFM reconstructions. The contrast was systematically increased through a reduction of the background noise level and the elimination of some artifacts. Finally, the proposed method led to a reduction of the computation time and file size by respectively a factor of 120 and 75.

Date	20 Dec 2022
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Pierre Bélanger (Supervisor) & Matthew Toews (Co-supervisor)