AUTHOR=Li Yanping , Li Honggang , Guan Yong , Zhang Xinyu , Zhao Xiaomei 

TITLE=Dense metal corrosion depth estimation

JOURNAL=Frontiers in Physics

VOLUME=Volume 11 - 2023

YEAR=2023

URL=https://www.frontiersin.org/journals/physics/articles/10.3389/fphy.2023.1277710

DOI=10.3389/fphy.2023.1277710

ISSN=2296-424X

ABSTRACT=Metal corrosion detection is important for protecting lives and property. X-ray inspection systems are widely used because of their good penetrability and visual presentation capability. They can visually display both external and internal corrosion defects. However, existing X-ray-based defect detection methods cannot present and estimate the dense corrosion depths. To solve this problem, we propose a dense metal corrosion depth estimation method based on image segmentation and inpainting. The proposed method employs an image segmentation module to segment metal corrosion defects and an image inpainting module to remove these segmented defects. It then calculates the pixel-level dense corrosion depths using the X-ray images before and after inpainting. Moreover, to address the difficulty of acquiring training images with ground-truth dense corrosion depth annotations, we propose a virtual data generation method for creating virtual corroded metal X-ray images and their corresponding ground-truth annotations. Experiments on both virtual and real datasets show that the proposed method successfully achieves accurate dense metal corrosion depth estimation.