Automatic Rebar Picking for Corrosion Assessment of RC Bridge Decks with Ground-Penetrating Radar Data

• Published in: Journal of performance of constructed facilities 2024-04, Vol.38 (2)
• Main Authors: Zhang, Yu-Chen, Du, Yan-Liang, Yi, Ting-Hua, Zhang, Song-Han
• Format: Article
• Language: English
• Subjects: Accuracy; Algorithms; Amplitudes; Bridge decks; Concrete bridges; Corrosion; Corrosion tests; Data processing; Ground penetrating radar; Hyperbolas; Image enhancement; Nondestructive testing; Radar data; Radar detection; Rebar; Structural safety
• ISSN: 0887-3828; 1943-5509
• DOI: 10.1061/JPCFEV.CFENG-4591

The detection and evaluation of internal corrosion of concrete bridge decks is of critical importance for bridge structure safety. Ground penetrating radar (GPR) is a nondestructive testing technique that has been used widely for corrosion assessment of RC bridge decks. However, the recognition accuracy of traditional automatic data processing methods for blurred hyperbolic features is insufficient, which leads to a decrease in the accuracy of bridge deck corrosion assessment. To address this problem, this paper proposes an automatic rebar picking algorithm based on gradient information and a migration method for bridge deck corrosion assessment with GPR data. This method can be divided into three steps: thresholding processing based on B-scan gradient information, a limited hyperbolic summation-based recognition (LHSR) algorithm, and rebar localization. First, the GPR B-scan is transformed into its gradient image and thresholded to enhance hyperbolic features. The LHSR algorithm then is applied to the binarized gradient B-scan to identify the hyperbolas, locate the rebar, and extract the rebar reflection amplitude. Finally, the corrosion map of the bridge deck is generated based on the rebar position and the rebar reflection amplitude after depth-correction. A case study with GPR data from two tested bridges was employed to validate the feasibility of the proposed method. The results show that the precision and recall of automatic rebar picking by this method for poor-quality GPR data were 91.27% and 93.56%, which are significantly higher than those of the traditional methods. Moreover, the accuracy of the bridge deck corrosion map obtained by the proposed method also is significantly better than that of the traditional methods. It can be concluded that the proposed method can be used for rebar picking and corrosion assessment of RC bridge decks with GPR data.