A hierarchical DCNN-based approach for classifying imbalanced water inflow in rock tunnel faces

Accurate water inflow assessment in the under-construction rock tunnel sites is critical for the next optimized construction and rehabilitation strategy. In this paper, a deep convolutional neural networks (DCNN)-based method, named H-ResNet-34, is implemented to classify water inflow category from rock tunnel faces in under-construction highway tunnels in Yunnan, China. An image database is compiled, which contains 8,000 images in five different water inflow categories of rock tunnel faces, namely complete dry (CD), wet state (WS), dripping state (DS), flowing state (FS) and gushing state (GS). Herein, a crucial issue is the imbalanced images between damage and non-damage owing to the vast sample of datasets and between various damages due to varying damage occurrence rates, which bring enormous challenges for conventional DCNN models. Thus, a hierarchical classification structure is applied to overcome the issue of imbalanced images at two different levels: coarse-level and fine-level. The coarse-level distinguishes the dataset with non-damage (i.e. complete dry) images. The fine-level computes the occurrence probability of the image dataset with water inflow damage. The constructed framework is then trained, validated, and tested using tunnel face images with various water inflow categories. The testing results suggest that the proposed hierarchical classifier is well competent for water inflow classification for rock tunnel face images and can effectively alleviate the imbalanced data issue.