Concrete sleeper crack detection based on image saliency

Abstract Concrete sleepers are crucial infrastructure in railway construction and operation. Sleeper cracks may cause the structural damage or functional failure of sleepers, endangering railway operation safety. A third-order residual crack saliency detection network (R 3 Net-CSD) model is constructed based on ConvNeXt for detecting concrete sleeper cracks in track inspection images. This model can accurately extract the saliency features of rail sleeper crack images. It first uses ConvNeXt to form the backbone network and a block module to form the first residual, followed by a spatial pyramid (dilated convolution group-based spatial pyramid) module based on the anti-residual structure and the dilated convolution group to form the second residual. Finally, it uses a saliency residual (significance residual) module to form the third residual. Simultaneously, combined with a multilevel joint loss function supervision strategy for supervised learning, the R 3 Net-CSD model fully utilizes the low-level edge information and high-level semantic information of cracks. By using the images collected by a hand-pushed rail inspection vehicle, a dataset of rail sleeper cracks is manually constructed and the developed model is experimentally validated. A large number of experiments demonstrate that the R 3 Net-CSD model achieves the best results in multiple indicators, including an average absolute error (mean absolute error) of 0.0082, an average accuracy (average precision) of 0.9764, and an area under the curve value of 0.9995. By combining the calculation method of crack geometric parameters, the quantitative indicators of crack length and width are obtained, with a detection accuracy of 0.1 mm. This research provides a new solution for crack detection in concrete sleepers and the measurement of geometric parameters.