Utilizing Deep Learning and Machine Vision in Tunneling: Classifying Rock Bodies, Assessing Perimeter Rock Quality, and Analyzing Groundwater Distribution

ABSTRACT As a crucial component of using subterranean space, tunnel engineering faces several difficulties with groundwater distribution analysis, rock quality assessment, and classifying rock masses. This work attempts to increase the precision of surrounding rock quality evaluation, the accuracy of rock structure classification, and the quantitative characterization of groundwater distribution in tunnel engineering by merging machine vision and deep learning approaches. Initially, we utilized the deep learning model Inception-ResNet-V2 to solve the challenge of classifying rock structures by automatically identifying and categorizing visible patterns in the rock body. The findings demonstrate that, when comparing alternative approaches, the Inception-ResNet-V2 model performs best when it comes to classifying samples of varied rock body structures. Second, we provide an integrated machine learning approach based on multi-source heterogeneous data to forecast enclosing rock categorization for the problem of enclosing rock quality evaluation. An accurate evaluation of the surrounding rock quality is attained by using the gradient boosted regression tree (GBRT) model in conjunction with the Bayesian optimization approach to optimize the hyperparameters. Lastly, we used DeepLabV3+, a deep learning semantic segmentation technique, to solve the groundwater distribution analysis problem and quantitatively quantify the groundwater distribution in the tunnel working face. The DeepLabV3+ model has good accuracy and stability in classifying groundwater flow patterns, as demonstrated by the experimental findings.