A survey on deep learning-based pavement crack segmentation methods

Pavement cracks, as common road distresses, play a critical role in road condition assessment, maintenance planning, and lifecycle management. Traditional manual inspections are labor-intensive and subjective, while conventional image processing methods are sensitive to lighting variations, background interference, and complex textures, limiting their practical applicability. Deep learning, with its powerful feature extraction and representation capabilities, has demonstrated clear advantages in complex environments and has become a major research focus for crack detection. This paper systematically reviews recent advances in deep learning-based pavement crack segmentation. It first introduces the task characteristics, commonly used datasets, and evaluation metrics. Then, it reviews methods based on convolutional neural networks and multi-scale feature fusion, lightweight models, attention mechanisms, Transformerbased approaches, and the latest Mamba state-space models. Finally, the paper summarizes existing challenges in pavement crack segmentation and discusses future research directions. This survey aims to provide a technical reference for the research and application of intelligent crack detection systems.