Presenter

Mr Mathias Becquaert - Royal Military Academy Brussels and ETRO, VUB [Email]

Abstract

Non-destructive testing (NDT) using cm- and mm-waves is an appealing technology, thanks to the attractive properties for in-depth imaging. The ultimate goal for this technology is creating high-resolution three dimensional images in depth, permitting the detection of defects, anomalies or targets hidden inside objects or in inaccessible areas. Those measurements induce expensive hardware, long image acquisition times and often generate large amounts of data.

An answer to these drawbacks can be given by applying the Compressed Sensing (CS) paradigm which promises an exact reconstruction of a sparse signal from only a fraction of the original data sampled at Nyquist rate, under certain conditions. This work is an in-depth study to prove the applicability and evaluate the performance of CS applied to Synthetic Aperture (SA) non-destructive measurements. A range of promising recent approaches are implemented, tested and evaluated on synthetically created and real data obtained through (1) a series of Through-the-Wall radar Imaging (TWI) experiments and (2) SA scans of 3D printed objects.

Through-the-wall radar images are a challenging application: Despite the fact that the scene is often populated by only a small number of targets, the images happen not to be sparse due to interaction between the sensing waves and the walls, contaminating the scene with low-rank signals. Several techniques exploiting the sparsity and the low-rank properties are explored in this work.

Additive manufacturing has become an extremely popular technology, with the use of three-dimensional printers, in many domains for prototyping, copying and producing complex objects. NDT techniques are needed to image the structure of these objects in order to identify defects or printing errors. One of the requirements of CS is that the reconstructed image must be sparse in a given domain. Often, this is not the only prior knowledge available: any meaningful Side Information (SI) could be added to CS minimization expression allowing an even more severe undersampling. In this work CS combined with homogeneous SI and CS with multimodal SI are tested and evaluated.


The defense takes place at the Royal Military Academy in Brussels, Hobbemastraat 8, 1000 Brussels
Conference hall in building I

Short CV

Civil Engineer-Polytechnicus, RMA, 2008