Graph Convolutional Network Detection Model for Pipeline Defects Based on Improved Label Graph

Baozhi Zeng; Jianqiao Luo; Ying Xiong; Bailin Li

doi:10.3788/LOP202259.1810004

Laser & Optoelectronics Progress, Volume. 59, Issue 18, 1810004(2022)

Graph Convolutional Network Detection Model for Pipeline Defects Based on Improved Label Graph

Baozhi Zeng, Jianqiao Luo, Ying Xiong, and Bailin Li^*

Author Affiliations

School of Mechanical Engineering, Southwest Jiaotong University, Chengdu 610031, Sichuan , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(9)

Fig. 1. ILG-GCN model framework

Download full size

Fig. 2. Network structure. (a) CNN network structure; (b) GCN network structure

Download full size

Fig. 3. Accuracy comparison of different $τ$ values

Download full size

Fig. 4. Accuracy comparison of different $k$ values

Download full size

Fig. 5. Classifier visual comparison.(a) Classifiers learned from CNN model; (b) classifiers learned from ILG-GCN model

Download full size

Fig. 6. Label graph comparison.(a) Label graph adopted by existing GCN model; (b) improved label graph adopted by ILG-GCN model

Download full size

Fig. 7. Comparison of model prediction results.(a) True label of sample;(b) prediction results of ML-GCN model;(c) prediction results of ILG-GCN model

Download full size

Table 1. Dataset partition
View table
Table 1. Dataset partition
Item Training set Validation set Testing set
Picture number 6668 2223 2223
Label number 3154 1067 1055

Table 2. AP and mAP of each model

View table

Table 2. AP and mAP of each model

Model	Deformation	Fracture	Leakage	Penetration	Obstacle	Root	Concrete	mAP
CNN	91.1	91.5	83.3	75.4	90.7	89.1	90.2	87.3
CNN-RNN	94.0	95.8	85.3	80.8	94.5	94.2	94.4	91.3
RAR	94.5	96.9	89.5	85.3	96.6	93.8	95.6	93.2
ML-GCN	94.1	95.8	90.9	86.6	96.8	97.5	95.8	93.9
CNN-GCN	93.9	96.0	91.1	87.0	97.1	97.9	95.9	94.1
VGG-Sewer	91.3	92.2	82.7	74.9	91.7	89.4	90.9	87.6
ILG-GCN	95.7	98.4	93.7	89.8	96.5	97.3	97.9	95.6

Tools

Get Citation

Copy Citation Text

Baozhi Zeng, Jianqiao Luo, Ying Xiong, Bailin Li. Graph Convolutional Network Detection Model for Pipeline Defects Based on Improved Label Graph[J]. Laser & Optoelectronics Progress, 2022, 59(18): 1810004

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites