Defects recognition in on-site GPR images of mountain tunnel linings based on YOLOv8N model

Ground Penetrating Radar (GPR) has been widely adopted for non-destructive detection of hidden defects behind tunnel linings in mountainous areas, including voids, insufficient compaction, and inadequate lining thickness. However, the manual interpretation of extensive B-scan images is time-consuming, labor-intensive, and heavily dependent on operators’ expertise and prior knowledge. This study employed the YOLOv8N model to detect rebar, void, and steel mesh for evaluating the quality of mountainous tunnel lining. A dedicated datasets comprising 1,030 labeled GPR images of tunnel linings, collected from real-world underground environments, was utilized to train and validate the model. Performance metrics, including precision, recall, and mean average precision (mAP), were employed to evaluate the model’s accuracy in defect identification and thickness estimation. Comparative experiments with Single Shot Multibox Detector (SSD) demonstrated the superior efficiency and reliability of the proposed algorithm, achieving an mAP of 91%. The results underscore the potential of deep learning-driven automation to enhance the objectivity and scalability of tunnel lining inspections, thereby supporting infrastructure maintenance and safety management.