Tunnels are essential underground or underwater infrastructure for setting up new communication routes overcoming significant orographic obstacles and/or geo-morphological constraints. The lining is the perimetral structural supporting part of a shaft excavation or a tunnel, and it consists of continuous support in unstable ground formations or even discrete limited supports in stable rock earth conditions made of steel or concrete, often with tie rods for stabilization purposes. Precisely, linings are composed of different layers, i.e. primary supports that aim to stabilize the opening in the ground excavation, permanent support demanded to carry long-term linings design loads, and temporary support installed only for short periods of time. Direct and indirect monitoring of the health state of tunnel lining is vital for guaranteeing adequate safety levels in time, promptly performing proper maintenance interventions, and extending the nominal life of these strategic infrastructures. Among several available existing techniques, indirect monitoring with ground-penetrating radar (GPR) become particularly attractive in the last decades because of the high productivity and rapidity of inspection sessions, which require tunnel traffic to shut down for very limited periods. The actual time-consuming part of the GPR-based monitoring is not the data collection procedure, but the collected profiles post-processing, which is still often a manual procedure nowadays. In this study, the authors proposed a computer vision (CV) artificial-neural-network-based solution for analyzing GPR profiles collected from the Italian road tunnels existing heritage for automatically detecting typical lining flaws. The proposed approach is based on a multi-level hierarchical binary classification set to intelligently categorize tunnel lining defects via the Vision Transformer (ViT) model.