AUTOMATIC DETECTION OF REFLECTIONS AT ANCIENT WALLS IN SIMPLE-OFFSET GPR IMAGES BY USING CASCADE CLASSIFIERS

MARTINELLI, PATRICIA; BORDÓN, PABLO; BONOMO, NÉSTOR

Porto

Congreso; Near Surface Geoscience Conference & Exhibition; 2018

European Association of Geoscientists and Engineerings

Simple-offset GPR reflectionmethodology allows obtaining very precise information inarchaeological/historical sites. However, as large amounts of data are usuallyacquired, their processing, analysis and interpretation can be extremelytime-consuming. In this work, we present threealgorithms for the automatic detection of reflections at ancient walls inSO-GPR images, based on cascade classifiers and well-known image featuredescriptors: Haar, HOG and LBP. These algorithms were implemented using supervised learning, andexperimental data from previous works. The bestperformances corresponded to the descriptor Haar. With only two cascade stages,remarkably accurate results were attained despite the complex characteristics of thesignals of the walls. Almost all of them were detected near their actualpositions, and only a few false positive predictions were obtained, mostlywithout any continuity across the profiles.  Themain advantage of these methodologies is that once an accurate and reliablealgorithm is implemented using data from an appropriate sector, it can beapplied in all the zones of the site with similar characteristics, or even inother site of the same type. Thereby, a precise representation of the targetstructures is rapidly obtained, and the qualified interpreter only has toexamine some parts of particular profiles.