Dataset for surface and subsurface characterization of removable urban pavements with a functionalized surface...

The evolution of transportation infrastructure has become an increasingly critical issue as extreme weather events intensify and traffic density grows, leading to a rise in visible surface degradation. In addition, mechanical stresses on pavements and potential infrastructure settlements have significant impacts on underground utility network nodes [1]. Regardless of the type of degradation, innovative solutions are required to design the roads of the future. In this context, an innovative system called Removable Urban Pavements (RUP) has been developed. Its purpose is to make easier and faster the access to the base layer and underground utility networks during maintenance operations. This modular structure can be opened and closed quickly, thereby reducing intervention time and preventing the longitudinal unevenness typically observed after trenching on standard pavements. The RUP structure consists of interlocking, prefabricated hexagonal slabs made of porous concrete with a draining surface [2], which helps prevent standing water and reduces the risk of hydroplaning. To characterize the condition of RUP and ensure reliable monitoring of their evolution, Non-Destructive Testing (NDT) methodology combining imaging and electromagnetic (EM) wave propagation has been implemented. Structural identification is performed using a drone (Unmanned Aerial Vehicle: UAV), enabling element-by-element detection and classification between RUP and standard pavements. Surface anomalies are detected from ground-level images captured with a smartphone acting as a stand-in for a robot (Unmanned Ground Vehicle: UGV) developed within the European HERON project [3]. Subsurface condition visualization is achieved using a Stepped-Frequency 3D Ground Penetrating Radar (GPR) system [4]. The ultimate goal is to develop an evolving database containing both structural components and the various detectable anomalies. A deep learning model based on a YOLO-type (You Only Look Once) architecture can then be used to perform automatic classification of these elements.