Employing vision transformers for crack detection and health monitoring of concrete structures

The safety and security of concrete structures is essential and should be regularly monitored by timely identifying deficiencies to avoid collapses which may lead to causalities and economic losses. The advancement in technology has enabled more automated flexible and smooth monitoring of concrete structures, including buildings, bridges, etc. Specialised cameras capture images which can be analysed for effective knowledge discovery. The work described in this paper addresses this serious issue by presenting a novel application of Vision Transformers (ViTs), a deep learning technique originally developed for image classification, to the task of crack detection in concrete structures. The main target is to improve crack and deficiency identification by utilising a thoroughly trained ViTs model using public and proprietary data sets. Cracks and damages in concrete structures are identified and classified with high accuracy. This has been illustrated by conducting extensive experiments which reported promising evaluation metrics values.