A review of deep learning methods for pixel-level crack detection

Cracks are a major sign of aging transportation infrastructure. The detection and repair cracks is the key to ensuring overall safety In recent years, due remarkable success deep learning (DL) in field crack detection, many researches have been devoted developing pixel-level image segmentation (CIS) models based on DL improve accuracy, but as far we know there no review DL-based CIS methods yet. To address this gap, present comprehensive thematic survey techniques. Our offers several contributions area. First, more than 40 papers journal or top conference most published last three years identified collected systematic literature method. Second, according backbone network architecture proposed them, they grouped into 10 topics: FCN, U-Net, encoder-decoder model, multi-scale, attention mechanism, transformer, two-stage multi-modal fusion, unsupervised weakly supervised learning, be reviewed. Meanwhile, our focuses discussing strengths limitations each topic so reveal latest research progress field. Third, publicly accessible data sets, evaluation metrics, loss functions that can used for systematically introduced summarized facilitate researchers select suitable components their own tasks. Finally, discuss six common problems existing solutions them CIS, then suggest eight possible future directions