CORROSION DETECTION FOR STEEL GIRDER BRIDGES USING A CONVOLUTIONAL NEURAL NETWORK WITH INFLUENCES THE NUMBER OF TRAINING DATA

Abstract

As the number of bridges to be inspected is increasing due to the aging of bridges and the number of inspection engineers is decreasing in recent days, the introduction of new technologies is an urgent issue. In recent years, the application of the convolutional neural network (CNN) as a machine learning has attracted attention as one of the effective inspection method in the civil engineering field. In a previous study, the learning model that could be used as a corrosion detector for steel girder bridges each construction office units was developed using photographs of road bridge inspection results from Fukushima Prefecture as the training data, which was reported to have the practical classification accuracy. In this study, we therefore investigated the possibility of improving the classification accuracy for the corrosion detection by developing the expanded learning model that added the training data of two areas from previous studies. As a result, the accuracy of detecting corosions with twice as the training data was higher by 8% than that with the training data of a single area.