DEEP LEARNING APPROACH FOR THE IDENTIFICATION OF STRUCTURAL LAYERS IN HISTORIC MONUMENTS FROM GROUND PENETRATING RADAR IMAGES

The present work is about the application of Artificial Intelligence and in particular Computer Vision approaches for the analysis and classification of Ground Penetrating Radar (GPR) B-Scan radargrams gathered during a GPR data acquisition campaign for the diagnostic study, for the assessment of the preservation state of the Holy Aedicule of the Holy Sepulchre in Jerusalem. The analysis of those data revealed the Aedicule’s structural layers and most important indicated the cause of the historical building pathology. The objective of this study is to extract the knowledge coming from the typical analysis of B-can radargrams, based on which the various structural layers derived, omitting this way several manual data pre-processing and time-consuming steps. The study employs a Deep Learning architecture, known as U-Net, where an image segmentation approach has been followed to build and train a classifier able to discriminate the various structural layers detected by the original measurements of radargrams.