Abstract
Bridge diseases, including steel corrosion, protective material flaking off, etc., lead performance deterioration of material and risk increasing of structure failure. To understand well where a disease is and which level it is as the stipulation of maintenance specification, lots of manual inspection works are performed every year. Manual visual inspection method is reliable and useful, but it is time-consuming and high-cost, also not suitable for a standardized tracking of a disease development. In this study, a computer vision system is developed for long-span steel box-girder bridge inspection. Correspondingly, an automatic framework of disease identification based on deep learning is proposed. Finally, an experiment of this system on Jiangyin Bridge is introduced. This systemized solution is a novel methodology to instead of manual visual inspection and it can be attached to an existed bridge inspection installation. The results indicate that it is reliable and has a high robustness in real scenarios.
