In concrete tunnel linings, cracks usually appear and develop as an early sign of structural degradation prior to severe intolerable serviceability damage. The monitoring and assessment of crack spatial distribution can highlight long-term tunnel structural behavior and facilitate tunnel maintenance. This study describes a remote and automated system for conducting crack monitoring at a pixel-level scale using robot-mounted imaging technology. This system collects crack images remotely and stitches them together to create a panorama image of the tunnel surface. Employing transfer learning, this study fine-tunes and improves the state-of-the-art semantic segmentation model with a lightweight backbone, DeepLab V3plus, to detect cracks automatically. A novel smooth blending prediction method is implemented on the panorama to present long-distance tunnel crack distribution, alleviating misclassification problems encountered in high-resolution image inference. In addition, transfer learning, tailored loss functions, and regularization techniques have been developed based on the CERN tunnel crack database characteristics to maintain high performance and generalization of the proposed method.Field trials conducted in tunnels at CERN demonstrate the feasibility of the proposed crack monitoring system. Results show that the proposed system allows the identification of severe crack-damaged tunnel sections and specific crack patterns, which can be related to the structural behavior of the tunnel lining.