Intelligent Detection of Tunnel Leakage Based on Improved Mask R-CNN

Symmetry, Journal Year: 2024, Volume and Issue: 16(6), P. 709 - 709

Published: June 7, 2024

The instance segmentation model based on deep learning has addressed the challenges in intelligently detecting water leakage shield tunneling. Due to limited generalization ability of baseline model, occurrences missed detections, false and repeated detections are encountered during actual detection tunnel leakage. This paper adopts Mask R-CNN as introduces a mask cascade strategy enhance quality positive samples. Additionally, backbone network is replaced with RegNetX enlarge model’s receptive field, MDConv introduced feature extraction capability edge field region. Building upon these improvements, proposed named Cascade-MRegNetX. MRegNetX features symmetrical block structure, which, when combined deformable convolutions, greatly assists extracting from corresponding regions. During dataset preprocessing stage, we augment through image rotation classification, thereby improving both quantity Finally, by leveraging pre-trained models transfer learning, robustness target model. can effectively extract areas different scales or deformations. Through experiments conducted comprising 766 images leakage, experimental results demonstrate that improved achieves higher precision detection. enhancements, effectiveness, capability, improved. Cascade-MRegNetX respective improvements 7.7%, 2.8%, 10.4% terms AP, AP0.5, AP0.75 compared existing Cascade

Language: Английский

---

Risk assessment of tunnel water inrush based on Delphi method and machine learning

Frontiers in Earth Science, Journal Year: 2025, Volume and Issue: 13

Published: March 21, 2025

The water inrush is one of the most catastrophic emergencies in metro tunnels. To avoid potential inrush, this paper proposes a risk assessment model for tunnel based on Delphi survey method and machine learning. proposed consists two parts, index system level prediction model. Firstly, by using method, appropriate factors are assembled into system. guarantee accuracy results, only correctly selected factors, validated Grey Relational Analysis (GRA), recognized as indexes. Then, Radial Basis Function (RBF) network, improved Locally Linear Embedding (LLE) algorithm Particle Swarm Optimization (PSO), applied to predict level. Training test sample sets constructed engineering data from Qingdao construction. In comparison with baseline models, demonstrates best mean square error, which 92.5% 0.015, respectively. LLE-PSO-RBF Metro Line 4 project. Three tunnels predicted invoking trained model, I, III IV,

Language: Английский

---

Image segmentation of tunnel water leakage defects in complex environments using an improved Unet model

Scientific Reports, Journal Year: 2024, Volume and Issue: 14(1)

Published: Oct. 16, 2024

Computer vision technology provides an intelligent means for detecting tunnel water leakage areas. However, the accuracy of defect feature extraction and segmentation is limited by factors such as insufficient lighting environmental interference inside tunnels. To address problem, this paper proposes a area network model called Customized Side Guided-Unet (CSG-Unet), using Unet baseline model. The main contributions are: (1) improve extraction, customized side guided term introduced to direct net's attention changes in light shade within image. A parallel module designed extract internal information from term. Subsequently, strengthened channel aggregates original achieve accurate areas; (2) scarcity datasets, basic dataset constructed collecting data open-source datasets manually gathered On basis, perspective transformation used change camera viewpoint, gaussian noise randomly added images simulate taken dimly lit scenes, thereby expanding enhancing network's generalization. CSG-Unet was trained training set, achieving mean Intersection over Union (mi IoU) 85.54%, Dice coefficient Dice) 85.26%, Pixel Accuracy PA) 90.85%. Compared its network, U-Net (tiny), these metrics show improvement 3.2% each indicator. Finally, visual comparison between improved further confirms that proposed can effectively adapt areas complex environments.

Language: Английский

---

Drilling Process Monitoring for Predicting Mechanical Properties of Jointed Rock Mass: A Review

Buildings, Journal Year: 2024, Volume and Issue: 14(7), P. 1992 - 1992

Published: July 1, 2024

Reliably assessing the quality and mechanical properties of rock masses is crucial in underground engineering. However, existing methods have significant limitations terms applicability accuracy. Therefore, a field measurement method that meets real-time monitoring safety requirements for engineering needed. Firstly, research findings domestic international scholars on application drilling process technology are comprehensively analyzed. Rotary cutting penetration tests conducted tuff containing fractures joints. Various mass classification evaluation standards integrated with rotary tests. used to determine residual strength rock, based this review. The rationality calculated mi parameter values validated. peak strength, errors obtained method. index designation from (RQDd) redefined, apparatus (DPMA). Rock conducted, correlation between standard deviation energy (RQD). Additionally, new relational formula introduced RQD variations energy, discontinuity frequency. This undoubtedly provides scientific basis design construction, ensuring long-term applications.

Language: Английский

---

Intelligent identification and semantic segmentation of deep rock fracture based on deep ensemble learning and transfer learning

Tunnelling and Underground Space Technology, Journal Year: 2024, Volume and Issue: 157, P. 106317 - 106317

Published: Dec. 25, 2024

Language: Английский

---