Autonomous detection of steel corrosion spatial variability in reinforced concrete using X-ray technology and deep learning-based semantic segmentation

Abstract

Correctly determining the spatial distribution of steel corrosion within a structural member is critical for estimating the remaining service life of deteriorating reinforced concrete (RC) structures. While X-ray technology serves as a nondestructive inspection method, existing challenges persist, particularly in semi-automated corrosion boundary detection. This paper describes a deep learning-based semantic segmentation framework to autonomously detect X-ray images associated with RC, facilitating the visualization of nonuniform steel corrosion distribution. X-ray images were collected from a comprehensive experiment using RC specimens with various structural details by two accelerated corrosion methods. Four deep learning models were constructed, trained, and compared based on the database containing the original X-ray images and the corresponding pixel-level labels. The results demonstrate that the proposed autonomous detection method can segment uncorroded steel at a very high level of global accuracy without time-consuming work, outperforming traditional methods in terms of both accuracy and efficiency.