Automated Tunnel Damage Report Generation

Assessing structural damage in ageing tunnels remains a major challenge due to the subjectivity and time-intensive nature of manual inspections. This paper presents a web-based automated framework that processes ultra-high-resolution panoramic tunnel images to generate detailed damage records and risk assessment reports. To tackle the issue of imprecise defect boundaries, we introduce a novel evaluation metric—Intersection over Union with buffer zone (IoUb)—which prioritizes overall damage representation over strict boundary accuracy. We benchmarked several instance segmentation algorithms and found that using a lower confidence threshold reduces missed detections while keeping false positives low. Experiments conducted on Italian road tunnels show that the framework effectively enhances damage detection, supports severity categorization, enables natural language queries for statistical insights, and integrates visualization and report generation. This end-to-end platform offers a streamlined and objective approach to tunnel condition assessment, paving the way for scalable and data-driven infrastructure maintenance.