Deep learning in crack detection: A comprehensive scientometric review

Abstract

Cracks represent one of the common forms of damage in concrete structures and pavements, leading to safety issues and increased maintenance costs. Therefore, timely crack detection is crucial for preventing further damage and ensuring the safety of these structures. Traditional manual inspection methods are limited by factors such as time consumption, subjectivity, and labor intensity. To address these challenges, deep learning-based crack detection technologies have emerged as promising solutions, demonstrating satisfactory performance and accuracy. However, the field still lacks comprehensive scientometric analyses and critical surveys of existing works, which are vital for identifying research gaps and guiding future studies. This paper conducts a bibliometric and critical analysis of the collected literature, providing novel insights into current research trends and identifying potential areas for future investigation. Analytical tools, including VOSviewer and CiteSpace, were employed for in-depth analysis and visualization. This study identifies key research gaps and proposes future directions, focusing on advancements in model generalization, computational efficiency, dataset standardization, and the practical application of crack detection methods.