Computer Vision and Code-Based Damage Detection in PRC and RC Bridges

This is the authors' abstract. We don't add key points for this paper.

This work presents an integrated methodology for the automated visual inspection of reinforced and prestressed concrete bridges, combining two complementary artificial intelligence-based procedures. The first focuses on the detection and classification of surface damage in structural elements such as beams, piers, and abutments, through a deep learning (DL) model trained to recognize both components and defect types in accordance with national inspection guidelines. The second procedure targets the identification and assessment of bridge bearings, detecting their typology and generating a preliminary evaluation based on visual evidence. Both methodologies rely on image datasets collected from real-world inspections and annotated using standardized taxonomies. The system delivers structured outputs, including defect categories, component localization, and confidence scores, allowing for consistent, rapid, and large-scale evaluation of bridge conditions. By reducing the subjectivity of manual inspections and supporting inspectors in the prioritization of maintenance actions, this approach offers a practical step forward toward scalable, AI-assisted bridge management. Moreover, the code-based defect classification helps managers reduce time-consuming activities like frequent bridge inspections and form compilation, allowing inspectors to focus more on complex or particularly deteriorated bridges.