AI for pavement distress detection on lower-ranking roads: A bibliometric and systematic review

Artificial intelligence (AI) and deep learning have transformed pavement condition assessment by enabling faster, more consistent, and scalable detection of pavement distresses compared with traditional manual inspection. Although significant progress has been made in detecting cracks and other common defects, limited attention has been given to developing AI-based approaches tailored to the specific needs of lower-ranking roads, where potholes, ruts, manholes, and localised surface irregularities occur more frequently and present substantial safety concerns. This paper presents a combined bibliometric and systematic review of recent advances in AI-driven pavement distress detection, with emphasis on methods applicable to low-volume road networks. A structured search of Scopus and ScienceDirect resulted in 98 relevant peer-reviewed studies. The bibliometric analysis revealed a strong increase in research activity since 2018, with China, India, and the United States of America leading the number of published articles. You Only Look Once (YOLO)-based object detection algorithms dominate current approaches, supported by variants of convolutional neural networks and emerging multi-modal techniques that incorporate unmanned aerial vehicle imagery, three-dimensional (3D) scanning, or sensor fusion. Despite rapid development in AI-driven pavement distress detection, key challenges persist, including inconsistent ground-truthing, limited comparison between AI and human evaluators, lack of standardised benchmarking datasets, and the need for robust 3D data acquisition strategies for complex distresses on lower-ranking roads. Addressing these gaps is essential to ensure reproducibility, comparability, and practical implementation of AI-based pavement inspection systems.