Automatic Corrosive Environment Detection of RC Bridge Decks from Ground-Penetrating Radar Data Based on Deep Learning

Abstract

Detecting the corrosive environment of reinforced concrete (RC) bridge decks is of critical importance for evaluating the reliability and safety of bridge structures. However, accurately and automatically detecting a corrosive environment with traditional methods is challenging. This paper proposes a method for the automatic corrosive environment detection of bridge decks from ground-penetrating radar (GPR) data based on the single-shot multibox detector (SSD) model. This method can be divided into three steps: data preprocessing, automatic rebar picking, and corrosive environment mapping. First, the GPR data are preprocessed to enhance the contrast of the hyperbolic feature in GPR B-scans. Then the rebars in the B-scan images are automatically picked up by the trained SSD model. Finally, the corrosive environment contour map of the bridge deck is generated with the rebar reflection amplitudes after depth correction. The SSD model was trained with 10,316 B-scan images and tested with 2,578 images. The 300×300-pixel B-scan image typically included three to five hyperbolas. A case study with GPR data from a tested bridge was employed to validate the feasibility of the proposed method. The results show that the accuracy of the automatic corrosive environment detection method can reach 98% and is considerably higher than that of commercial software methods.