Automated Damage Report Generation via Panoptic Segmentation of Ultra-High-Resolution Tunnel Images

Modern reality capture techniques, such as ultra-high-resolution (UHR) panoramic imaging, can digitally document tunnel surface conditions in detail. However, assessing severe structural damage in aging tunnels still depends on manual inspections—an inefficient and subjective approach. This gap between advanced data collection and practical risk assessment limits effective tunnel maintenance. In this paper, we propose a novel framework that uses UHR panoramic images to automatically detect damage, generate records and risk assessment reports. At its core, a flexible segmentation framework, enhanced by a side network, captures contextual information around each local patch for sufficient receptive field. This method enables direct panoptic segmentation of images over 6k resolution, accurately identifying auxiliary structures and five types of damage. The proposed framework significantly improves the efficiency of planning and monitoring aging tunnel assets