Image-based crack identification for concrete bridges using region-based convolutional neural network

Detecting concrete cracks is required in current condition assessment of bridges. In this study, a faster region-based convolutional neural network (faster R-CNN) was applied for image-based crack identification, where the real-world images taken from concrete bridges were contaminated with complex backgrounds including handwriting. For this purpose, dataset of 2489 cropped images was created and labelled for two different objects, i.e. cracks and handwriting. The proposed network was then trained and tested using the generated image dataset. Full-scale real-world images were then used to evaluate the performance of the trained network. The results of this study show that the faster R-CNN can locate crack automatically from raw images, even with the presence of handwriting scripts.