Ground penetrating radar hyperbola detection using Scale-Invariant Feature Transform

Ground-penetrating radar (GPR) is a geophysical method used for subsurface targets identification. However, a sophisticated processing of GPR data is necessary to extract targets signature. Traditionally, we use the Hough Transform to detect the hyperbolas caused by targets reflections. This one is very difficult in implementation because we deal with a three-dimensional parametric space. In fact, the Hough transform consume a huge execution time and gives random results. In a previous work, we proposed a modified Hough Transform algorithm, based on a pre-treatment phase of the B-Scans images, that give acceptable results but its a little time consuming. This paper proposes an application of one modern computer vision feature extraction technique, named Scale Invariant Feature Transform (SIFT), to hyperbola detection in GPR Bscans data. First, we will apply the SIFT algorithm to localize the hyperbolas regions, and then we apply the Hough transform on the detected regions. The results presented indicates that SIFT features provide a robust tool for target identification for both classification and prescreening.