ABSTRACT
In many countries, thousand kilometers of underground tunnel infrastructure require regular inspection and monitoring for structural condition assessment and asset maintenance. In current practice, gathered field monitoring data are usually stored and processed by individual asset stakeholders for specific tasks exclusively, but not integrated to one database openly available for wider engineering society. This study addresses this gap by developing a Tunnel ImageNet database for enhanced tunnel structural assessment utilizing machine learning. The aim is to create a robust, standardized, well-annotated and structured image database for classifying diverse tunnel defects such as structural damage, material degradation, facility defects, etc. The image data will be structured and organized based on principles derived from WordNet synsets and metadata labeling to embed efficient hierarchical properties within images. Additionally, data storage and management will be handled, ensuring secure, interoperable, and easily accessible data repositories. The compiled Tunnel ImageNet database will be made openly available for training and testing deep learning models in object classification tasks pertinent to condition assessment of tunnel infrastructure.
