A comprehensive deep learning methodology for the detection and classification of road pavement distress

According to the 2025 report of the European environmental agency, noise associated with road infrastructures represents the main source of acoustic pollution in urban areas. Since the detrimental impact on human health associated with a prolonged exposure to high levels of noise are well documented, the development of effective monitoring and mitigation techniques for road noise represent a priority in the field of environmental acoustics. The acoustic emission associated with tire-road interaction represents one of the main components of the overall noise in urban areas and it is strongly correlated, for frequencies below 1 kHz, to the texture of the road pavement. Traditional pavement evaluation techniques usually require lengthy procedures, which have to be carefully supervised by specialized personnel. A growing interest is being dedicated by the new possibilities for automatic classification disclosed by the recent advancements in the field of deep learning. In particular, approaches based on computer vision and time series identification might represent valuable resources for the indirect analysis of road pavement conditions. Furthermore, a recently proposed acoustic index, called Tire cavity noise. might provide valuable insight on both the acoustic emission and the texture of road pavements. The present study aims at assessing the feasibility of a synergic system for the monitoring of road pavement health that could integrate information provided by tire cavity noise sensors and computer vision algorithms; in order to provide a comprehensive and thorough characterization of the road pavement. First, the technological challenges associated with the development of a TCN sensor were addressed and a functioning prototype was proposed. Next, computer vision algorithms for the purpose of road surface monitoring were evaluated. In the present study, a U-shaped neural network was selected for the development of an automatic crack segmentation system. Finally, a mobile laboratory, equipped with the previously developed instrumentation, was used to perform real-life measurements,in order to assess the feasibility of a synergic approach integrating tire cavity noise analysis and computer vision algorithms. The results of the case study highlighted the limitations associated with the current approach and led to the proposal of a new synthetic index for the evaluation of road pavement conditions.