Intelligent tunnel lining defect detection: advances in image acquisition and data-driven techniques

Abstract Tunnel environments often posed challenges such as complex backgrounds, low lighting, and low contrast, while the availability of open-source tunnel defect image data remained limited. Data augmentation techniques emerged as crucial methods to address these issues and enhance model generalization. This paper systematically reviews 276 key publications from 2018 to 2024, providing a comprehensive overview of the latest research progress, particularly in image acquisition and data augmentation, for intelligent tunnel lining defect detection. It began by introducing various methods for capturing images of tunnel surface and internal defects, including digital photography, laser scanning, and ground-penetrating radar (GPR) techniques, while analyzing their respective advantages and limitations. The discussion then focused on critical aspects of constructing defect datasets, such as image processing, data annotation, and the availability of public datasets, highlighting challenges associated with data collection and labeling. Furthermore, this study summarized the major challenges faced in the field, including high costs of data collection and annotation, a lack of diverse and comprehensive datasets, and the computational resource demands of advanced augmentation methods. Based on these challenges, the paper proposed future research directions, including the acquisition of more real-world GPR data, the development of public tunnel defect datasets, and the exploration of lightweight data augmentation techniques. These directions aimed to enhance the robustness and generalization of tunnel defect detection models. They also aimed to improve the efficiency and practicality of these models for real-world applications. This comprehensive review serves as a valuable reference for researchers and practitioners. It is especially useful for those engaged in intelligent infrastructure inspection and maintenance using advanced computer vision techniques.