Abstract:The structural performance of bridges gradually deteriorates with increasing service life and the development of defects, affecting the safe operation of structures. The traditional bridge inspection primarily relies on the manual visual assessment, which suffers from low efficiency, strong subjectivity and traffic disruption, highlighting the need for efficient and objective evaluation of bridge exterior conditions. Therefore, an intelligent detection method integrating UAV aerial photography, deep learning and image analysis technology is proposed. For defect identification, the YOLOv11-seg instance segmentation model is optimized by introducing three attention mechanisms (MaSA, CGA and EMA). A self-built dataset covering cracks and spalling defects is used with data augmentation applied to enhance sample diversity. Size quantification combined with morphological methods and reference object calibration techniques can realize the automatic measurement of crack length, width and spalling area. The result shows that the model integrating the EMA mechanism is greatly improved in recognition precision, recall rate and F1-score, enabling accurate defect identification. The size quantification maintains a relative error within 10%, meeting the engineering accuracy requirements. The validation through real bridge cases demonstrates that the mentioned intelligent detection system can achieve the key functions such as defect identification, size extraction and defect localization, reflecting the good engineering practicability and promotion value.
