China Journal of Highway and Transport

Intelligent Recognition Method of Tunnel Face Joints and Fissures Using Convolutional Neural Network

Yun-bo ZHANG, Ming-feng LEI, Yong-zhuo XIAO, Guang-hui LIU, Xing-xing DENG, Fu-yu YANG, Bao-jin LU, Chong-yang LI

China Journal of Highway and Transport, 2024, 37(7): 35-45. DOI: 10.19721/j.cnki.1001-7372.2024.07.003

节理裂隙序号	骨架像素量	单位像素长度/m	节理裂隙识别测量长度/m	节理裂隙现场实测长度/m
1	498	0.003 33	1.66	1.72
2	126		0.42	0.45
3	901		3.00	3.18
4	805		2.68	2.72
5	90		0.30	0.33
6	277		0.92	1.00
7	210		0.70	0.75
8	125		0.42	0.44
9	138		0.46	0.51
10	130		0.43	0.45
11	333		1.11	1.20
12	588		1.96	2.09

Table 5 Mask Skeleton Dimension Statistics

Other figure/table from this article

Fig. 1 Mask R-CNN Network Structure
Fig. 2 Structure Diagram of ResNet101 and FPN
Fig. 3 Comparison of RoIAlign and RoIPool
Fig. 4 Structure Diagram of Mask R-CNN-E
Fig. 5 MBConv and SE Network Architecture Diagram
Fig. 6 Typical Tunnel Face Images
Fig. 7 Annotation Method of Images
Fig. 8 Ways of Data Augmentation
Fig. 9 Training Error Curves
Fig. 10 Training Accuracy Curves
Table 1 Model Performance Evaluation Metrics
Table 2 Single-class Evaluation Metric A₁ for Mask R-CNN-E and Mask R-CNN
Fig. 11 Recognition Effect of Joints and Fissures in the Complete Tunnel Face
Table 3 Comparison of Image Recognition of Partially Cropped Tunnel Face Under Different Algorithms
Table 4 Coordinates of the Typical Tunnel Face Image Detection Frame and the Amount of Mask Pixels
Fig. 12 Display of Different Joints and Fissures Masks Individually
Fig. 13 Mask Skeletonization
Fig. 14 Intelligent Identification and Measurement of Tunnel Face Joint and Fissure Compared with Actual Field Measurement Lengths