Bridge crack image classification based on improved transformer

As a major bridge country, China's facilities are vital to the transportation network and regional economic development. However, bridges exposed to the natural environment and traffic pressure for a long time are prone to problems such as cracks and corrosion, which, if not detected and repaired promptly, may lead to structural failures, affecting safety and economic benefits. Traditional image classification methods are susceptible to interference from shadows and dirt in practical applications, leading to reduced classification accuracy and limitations in dealing with fine-grained features and small sample datasets. In this paper, based on the Transformer model, combined with AlignMixup and Omage2Token data enhancement methods, the accuracy and recognition ability of the model on fine-grained and small-sample image classification is improved.