Compact Convolutional Transformer Fourier analysis for GPR tunnels assessment

Road tunnels are critical infrastructures that require regular maintenance and inspection to ensure the safety of users. However, traditional inspection methods can be costly, time-consuming, and may not provide an objective assessment of the tunnel's condition. To overcome these challenges, researchers have recently developed newly promising approaches that combine tunnel's indirect testing, e.g. ground-penetrating radar (GPR), with the potentialities of artificial intelligence (AI). This innovative technique allows for the automatic detection and classification of defects in tunnel linings, providing a faster, more efficient, and more reliable method for assessing the health of road tunnels. This study explores the application of one of the state-of-the-art deep learning models, i.e. the compact convolution transformer (CCT), to classify defects in GPR images of tunnel linings. The dataset used in this study consists of inspections of tunnels built between the 1960s and 1980s. Two CCT models were trained on filtered and unfiltered datasets and then compared to assess the effects of noise on the identification task.