Ultrasonic Guided Wave Dataset for Super-Resolution Reconstruction of Images from Compressed Wavefield Acquisition

Abstract

Full-wavefield ultrasound imaging is an inspection methodology which provides accurate diagnosis of defects in structures. However, full-wavefield acquisition and processing is often a time consuming process. To tackle this problem, Compressive Sensing (CS) reconstruction algorithms have been proposed to speed up the acquisition process. Such reconstruction can be combined with Super-resolution Convolutional Neural Networks (SRCNN) schemes to obtain high-resolution images from low-resolution wavefield images. This paper proposes the combination of CS and SRCNN to recover images captured with a Scanning Laser Doppler Vibrometer (SLDV), by training the SRCNN with a dataset of ultrasound wave propagation images captured on typical aerospace structures using the SLDV. For the training of the SRCNN, a dataset of 300 structural full-wavefield images were captured on four different aerospace panels and various simulated defects. CS sub-sampling procedures were used to generate a dataset of simulated low-resolution image acquisition. The SRCNN was then trained to recover the original high-resolution images from the low-resolution images obtained after CS sub-sampling. The results demonstrate the capability of the technique for enhancing image resolution and quality while acquiring just 20% of the original number of scan points.