Identification Methods for Structural Problems of Bridges Based on Deep Convolutional Neural Network

Faster R-CNN), structural problems of bridges, identification network model The safety of bridges, which are key components of roads, has attracted the attention of experts, technical engineers, and maintenance managers.Various apparent problems such as cracks, voids and pits, white precipitate, and corrosion must be identified during the visual inspection of a bridge.In this study, using soft-nonmaximum suppression (NMS), we improved the original NMS algorithm of the faster regional convolutional neural network (Faster R-CNN) and built a fine identification network model to classify and identify the structural problems of bridges to effectively reduce the missed detection rate.In addition, through manual inspection by photography, the average accuracies of the identification of problems, namely, voids and pits, cracks, and white precipitate, can reach 80.86, 81.42, and 87.39%, respectively, which are about 20 percentage points higher than that of the original Faster R-CNN.