Fine-grained Classification of Sleeper Shoulder Crack Images Based on Improved B-CNN

Qinan Li; Haixin Sun; Kejia Sun

doi:10.3788/LOP57.141013

Laser & Optoelectronics Progress, Volume. 57, Issue 14, 141013(2020)

Fine-grained Classification of Sleeper Shoulder Crack Images Based on Improved B-CNN

Qinan Li, Haixin Sun^*, and Kejia Sun

Author Affiliations

School of Electronic and Information Engineering, Lanzhou Jiaotong University, Lanzhou, Gansu 730070, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(11)

$Crack images of bi-block sleeper shoulder. (a) Normal; (b) invisible crack; (c) obvious crack; (d) fracture crack$

Fig. 1. Crack images of bi-block sleeper shoulder. (a) Normal; (b) invisible crack; (c) obvious crack; (d) fracture crack

Download full size

Fig. 2. B-CNN model structure

Download full size

Fig. 3. B-CNN forward operation mode. (a) No sharing; (b) partial sharing; (c) full sharing

Download full size

Fig. 4. Improved B-CNN model structure

Download full size

Fig. 5. Crack images of concentrated sleeper shoulder in test set. (a) Invisible crack sleeper; (b) obvious crack sleeper

Download full size

Fig. 6. Classification accuracy curves of training set

Download full size

Fig. 7. Loss rate curves of training set

Download full size

Table 1. Comparison of accuracy and feature dimensions of different models under full sharing mode on verification set and test set
View table
Table 1. Comparison of accuracy and feature dimensions of different models under full sharing mode on verification set and test set
Model Verification set accuracy /% Test set accuracy /% Feature dimension
B-CNN 90.48 91.34 16×16×512
B-CNN_GAP 91.62 92.47 1×1×512
B-CNN_GMP 90.63 90.48 1×1×512
B-CNN_GAMP 91.48 90.06 1×1×1024

Table 2. Fine-grained classification results of test set

View table

Table 2. Fine-grained classification results of test set

Category	FPR /%			FNR /%
Category	VGG-D	B-CNN	Improved B-CNN	VGG-D	B-CNN	Improved B-CNN
Normal	3.85	3.23	2.28	14.13	9.04	6.74
Invisible crack	5.35	3.74	2.83	9.26	7.69	7.47
Obvious crack	4.29	3.67	3.05	17.28	13.44	8.89
Fracture crack	2.42	1.13	0.94	5.99	4.65	4.07

Table 3. Classification results of three models on verification set and test set
View table
Table 3. Classification results of three models on verification set and test set
Model verification set accuracy /% Test set accuracy /% Recall /% Precision /% F1 /%
VGG-D 87.78 88.07 88.10 88.34 88.22
B-CNN 91.89 91.19 91.25 91.29 91.27
Improved B-CNN 93.89 92.76 92.80 92.77 92.79

Table 4. Comparison of feature extraction speed and model parameters of different models
View table
Table 4. Comparison of feature extraction speed and model parameters of different models
Model Feature extractionspeed /(frame·s^-1) Parametersize /MB
VGG-D 0.64 949.83
B-CNN 0.94 175.81
Improved B-CNN 0.93 175.83

Tools

Get Citation

Copy Citation Text

Qinan Li, Haixin Sun, Kejia Sun. Fine-grained Classification of Sleeper Shoulder Crack Images Based on Improved B-CNN[J]. Laser & Optoelectronics Progress, 2020, 57(14): 141013

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Image Processing

Received: Oct. 18, 2019

Accepted: Dec. 11, 2019

Published Online: Jul. 28, 2020

The Author Email: Haixin Sun (1402957265@qq.com)

DOI:10.3788/LOP57.141013

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Comparison of accuracy and feature dimensions of different models under full sharing mode on verification set and test set

Table 1. Comparison of accuracy and feature dimensions of different models under full sharing mode on verification set and test set

Table 2. Fine-grained classification results of test set

Table 2. Fine-grained classification results of test set

Table 3. Classification results of three models on verification set and test set

Table 3. Classification results of three models on verification set and test set

Table 4. Comparison of feature extraction speed and model parameters of different models

Table 4. Comparison of feature extraction speed and model parameters of different models