Advanced Rock Type Classification Using Deep Learning: A Data-Driven Approach to Tunnel Stability Monitoring and Maintenance

초록

Tunnels serve a pivotal role in various infrastructure systems-including roads, railways, and underground facilities-and ensuring their stability is of critical economic and societal importance. Traditional rock classification methods often rely on on-site geological surveys and petrographic/chemical analyses, which can be time-consuming, cost-intensive, and prone to subjective judgments. With the recent advancement in big data and artificial intelligence (AI) technologies, particularly Deep Learning, it has become possible to automatically learn and classify intricate rock patterns from large-scale image datasets. In this study, we propose a deep learning-based rock classification framework and apply several state-of-the-art models- ResNet, EfficientNet, MobileNet, and others-to both broad categories (igneous, sedimentary, metamorphic) and more fine-grained rock types (e.g., marble, sandstone, phyllite). Our results show that while high accuracies exceeding 90% can be achieved for three-class problems, classification performance drops significantly when expanded to 18 or more subclasses, likely due to the complex microstructures and ambiguous boundaries among rock types. These findings highlight the substantial potential of deep learning in tunnel stability monitoring and maintenance, yet also underscore the necessity of hierarchical classification strategies, multimodal data integration (e.g., microscopic or hyperspectral imaging), and further hyperparameter optimization for more refined rock classification tasks.

키워드