Abstract
Corrosion is one of the most common and severe problems in the daily maintenance of steel bridges. So far manual visual inspection remains a popular method in the corrosion detection, which means a large amount of labor power as well as material and financial resources are wasted in routine maintenance and quantitative criteria of evaluating corrosion could be hardly obtained due to the subjective judgments of inspectors. With the development of computer vision techniques in recent years, corrosion inspection could be achieved through image segmentation based on deep neural networks besides quantitative indexes such as area, distribution and corrosion rate could be obtained at the same time, which quantifies and standardize the evaluation criteria of corrosion in steel bridges. Furthermore, image segmentation and analysis based on deep neural net-work shows high robustness in the real complex scenarios and could be applied in actual bridge inspection rather than laboratory scenarios.
