92 / 2021-11-30 18:47:35

Autonomous water leakage detection in rock tunnel face images using deep learning techniques

faster region-based convolutional neural network; image detection; deep learning; water inflow.

Intelligent Equipment and Technology > 6. Cloud Computing and Big Data Analysis for Mines

Detecting and assessing the water inflow on under-construction rock tunnel sites to improve productivity and safety has formed an integral part of computer vision (CV)-based research in infrastructure engineering. Nevertheless, conventional CV methods are explored for automated interpretation of target images, which requires excessive image pre-processing and complex feature extractor. As such, a novel approach, known as faster region-based convolutional neural network (Faster R-CNN), was proposed for automatically detecting water inflow state, including, dry state (DS), wet state (WS), flowing state (FS), and gushing state (GS). The detection framework was trained, validated and tested by feeding 5,320 images (with 648×691 pixels) collected from under-construction rock tunnel face of the Mengzi-Pingbian High Speed project in Yunnan, China. Then, the trained model was evaluated in terms of computation cost and detection accuracy by applying detection speed, training time, missing rate, and mean average precision (mAP), respectively. The results show that the proposed method can indeed detect water leakage damages accurately and efficiently. This study lays the foundation for employing deep learning (DL) methods in water leakage damage detection meanwhile addressing similar issues for construction management.