Data Augmentation of Raman Spectral and Its Application Research Based on DCGAN

Table 1. CNN network design
View table
View in Article
Table 1. CNN network design
网络层设置参数
INPUT 预处理后的拉曼光谱数据
Conv1 Size: 1×5, stride: 1, ReLU
Conv2 Size: 1×5, stride: 1, ReLU
Full Connected Conv2的feature展开
OUTPUT 9个输出神经元, 连接FC层

Table 2. Generator network and Discriminator network for Raman spectral augmentation

View table

View in Article

Table 2. Generator network and Discriminator network for Raman spectral augmentation

生成网络						判别网络
网络层	卷积核	stride	padding	激活函数	BN层	网络层	卷积核	stride	padding	激活函数	BN层
deconv1	1×4	0	0	ReLU	是	conv1	1×5	2	1	LeakyReLU	是
deconv2	1×5	2	1	ReLU	是	conv2	1×4	2	1	LeakyReLU	是
deconv3	1×4	2	1	ReLU	是	conv3	1×4	2	1	LeakyReLU	是
deconv4	1×4	2	1	ReLU	是	conv4	1×4	2	1	LeakyReLU	是
deconv5	1×4	2	1	ReLU	是	conv5	1×5	2	1	LeakyReLU	是

Table 3. Distribution of corresponding drugs in data set National Institute for Food and Drug Control
View table
View in Article
Table 3. Distribution of corresponding drugs in data set National Institute for Food and Drug Control
序号药品类别样本数目
1 Gatifloxacin 15
2 Lomefloxacin 21
3 Norfloxacin 15
4 Pefloxacin 12
5 Cephradine 18
6 Cefradline 15
7 Cefixime 9
8 Ceftazidime 18
9 Cefdinir 30

Table 4. The training set, test set distribution of drug samples
View table
View in Article
Table 4. The training set, test set distribution of drug samples
药品训练集测试集
Gatifloxacin 11 4
Lomefloxacin 15 6
Norfloxacin 11 4
Pefloxacin 9 3
Cephradine 13 5
Cefradline 11 4
Cefixime 7 2
Ceftazidime 13 5
Cefdinir 21 9
总数 111 52

Table 5. Detailed results of Raman spectrum discrimination of China food and drug institute-SVM (%)
View table
View in Article
Table 5. Detailed results of Raman spectrum discrimination of China food and drug institute-SVM (%)
药品分错数量
训练集测试集
Gatifloxacin 5 3
Lomefloxacin 4 2
Norfloxacin 4 2
Pefloxacin 7 3
Cephradine 4 3
Cefradline 5 3
Cefixime 6 2
Ceftazidime 4 4
Cefdinir 2 3
分类准确率 63.06(70/111) 51.92(27/52)

Table 6. Detailed results of Raman spectrum discrimination CNN (%)
View table
View in Article
Table 6. Detailed results of Raman spectrum discrimination CNN (%)
药品分错数量
训练集测试集
Gatifloxacin 3 2
Lomefloxacin 2 1
Norfloxacin 2 2
Pefloxacin 4 2
Cephradine 2 1
Cefradline 3 2
Cefixime 5 2
Ceftazidime 2 1
Cefdinir 1 0
分类准确率 78.38(87/111) 75.00(39/52)

Table 7. The training set, test set distribution of drug samples
View table
View in Article
Table 7. The training set, test set distribution of drug samples
药品训练集测试集
Gatifloxacin 70 30
Lomefloxacin 70 30
Norfloxacin 70 30
Pefloxacin 70 30
Cephradine 70 30
Cefradline 70 30
Cefixime 70 30
Ceftazidime 70 30
Cefdinir 70 30
总数 630 270

Table 8. Detailed results of Raman spectrum discrimination of China food and drug institute-Data augmentation and CNN (%)
View table
View in Article
Table 8. Detailed results of Raman spectrum discrimination of China food and drug institute-Data augmentation and CNN (%)
药品分错数量
训练集测试集
Gatifloxacin 9 1
Lomefloxacin 5 3
Norfloxacin 8 2
Pefloxacin 10 2
Cephradine 4 2
Cefradline 8 4
Cefixime 7 1
Ceftazidime 8 3
Cefdinir 7 4
分类准确率 89.52(564/630) 91.85(248/270)

Table 9. LVE signal to noise ratio of augmented spectral by slope-bias adjusting (corresponding to Fig.3)
View table
View in Article
Table 9. LVE signal to noise ratio of augmented spectral by slope-bias adjusting (corresponding to Fig.3)
原始谱图(a) 扩增谱图(b)
24.56 31.24
29.74
30.18
32.05
31.77
34.65
28.47
30.69
29.69
31.51

Table 10. Detailed results of Raman spectrum discrimination of DCGAN and CNN (%)
View table
View in Article
Table 10. Detailed results of Raman spectrum discrimination of DCGAN and CNN (%)
药品分错数量
训练集测试集
Gatifloxacin 3 1
Lomefloxacin 4 0
Norfloxacin 2 2
Pefloxacin 3 0
Cephradine 1 0
Cefradline 3 0
Cefixime 2 0
Ceftazidime 3 0
Cefdinir 2 1
分类准确率 96.35 98.52

Table 11. LVE signal to noise ratio of augmented spectral by DCGAN (corresponding to Fig.4)
View table
View in Article
Table 11. LVE signal to noise ratio of augmented spectral by DCGAN (corresponding to Fig.4)
原始谱图(a) 扩增谱图(b)
27.67 29.28
27.08 30.59
28.65 31.63
20.49 31.79
29.01 32.02
27.53 31.67
20.65 29.28
28.22 29.32
21.56 30.37
20.86 29.34

Tools

Get Citation

Copy Citation Text

Ling-qiao LI, Yan-hui LI, Lin-lin YIN, Hui-hua YANG, Yan-chun FENG, Li-hui YIN, Chang-qin HU. Data Augmentation of Raman Spectral and Its Application Research Based on DCGAN[J]. Spectroscopy and Spectral Analysis, 2021, 41(2): 400

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Research Articles

Received: Feb. 5, 2020

Accepted: Jun. 2, 2020

Published Online: Apr. 8, 2021

The Author Email: LI Yan-hui (1703201023@mails.guet.edu.cn)

DOI:10.3964/j.issn.1000-0593(2021)02-0400-08

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. CNN network design

Table 1. CNN network design

Table 2. Generator network and Discriminator network for Raman spectral augmentation

Table 2. Generator network and Discriminator network for Raman spectral augmentation

Table 3. Distribution of corresponding drugs in data set National Institute for Food and Drug Control

Table 3. Distribution of corresponding drugs in data set National Institute for Food and Drug Control

Table 4. The training set, test set distribution of drug samples

Table 4. The training set, test set distribution of drug samples

Table 5. Detailed results of Raman spectrum discrimination of China food and drug institute-SVM (%)

Table 5. Detailed results of Raman spectrum discrimination of China food and drug institute-SVM (%)

Table 6. Detailed results of Raman spectrum discrimination CNN (%)

Table 6. Detailed results of Raman spectrum discrimination CNN (%)

Table 7. The training set, test set distribution of drug samples

Table 7. The training set, test set distribution of drug samples

Table 8. Detailed results of Raman spectrum discrimination of China food and drug institute-Data augmentation and CNN (%)

Table 8. Detailed results of Raman spectrum discrimination of China food and drug institute-Data augmentation and CNN (%)

Table 9. LVE signal to noise ratio of augmented spectral by slope-bias adjusting (corresponding to Fig.3)

Table 9. LVE signal to noise ratio of augmented spectral by slope-bias adjusting (corresponding to Fig.3)

Table 10. Detailed results of Raman spectrum discrimination of DCGAN and CNN (%)

Table 10. Detailed results of Raman spectrum discrimination of DCGAN and CNN (%)

Table 11. LVE signal to noise ratio of augmented spectral by DCGAN (corresponding to Fig.4)

Table 11. LVE signal to noise ratio of augmented spectral by DCGAN (corresponding to Fig.4)