Spectroscopy and Spectral Analysis, Volume. 42, Issue 3, 769(2022)
Detection of Pearl Powder Adulteration Based on Raman Spectroscopy and DCGAN Data Enhancement
Figures & Tables(10)
Fig. 1. Mean Raman spectra of the samples (a): Original spectra; (b): Spectra with pretreatment
Fig. 4. Original spectrum and generated spectrum based on DCGAN
(a): Original spectra; (b): Generated spectra
Fig. 5. Correlation curve between real and predicted purity of quantitative models built by different data enhancement methods combined with 1DCNN
(a): DCGAN-1DCNN; (b): Noise addition-1DCNN; (c): Translation-1DCNN; (d): Noise+Translation-1DCNN
Table 1. Parameters of generate network
View tableView in ArticleTable 1. Parameters of generate network
网络层 1维卷积核 步长 Padding 激活函数 Conv1 (16, 9) 3 same ReLU Conv2 (8, 18) 3 same ReLU Conv3 (3, 27) 3 same ReLU
Table 2. Parameters of discriminating network
View tableView in ArticleTable 2. Parameters of discriminating network
网络层 1维卷积核 步长 Padding 激活函数 Conv1 (3, 27) 3 same LeakyReLU Conv2 (8, 18) 3 same LeakyReLU Conv3 (16, 9) 3 same LeakyReLU
Table 3. Similarity evaluation between the spectra generated by traditional data enhancement and DCGAN enhancement methods and the original spectra
View tableView in ArticleTable 3. Similarity evaluation between the spectra generated by traditional data enhancement and DCGAN enhancement methods and the original spectra
样本
纯度/%左右平移 叠加噪声 平移+噪声 DCGAN PSNR/dB SSIM PSNR/dB SSIM PSNR/dB SSIM PSNR/dB SSIM 100 23.57 0.919 9 32.83 0.581 1 24.03 0.620 3 47.67 0.973 7 95 23.99 0.929 2 31.49 0.495 2 24.22 0.623 0 54.67 0.992 6 90 22.15 0.889 7 30.19 0.450 8 24.84 0.567 0 46.58 0.994 7 85 24.18 0.916 3 29.10 0.416 7 23.88 0.527 6 51.61 0.996 1
Table 4. Comparison of the identification results of adulterated pearl powder
View tableView in ArticleTable 4. Comparison of the identification results of adulterated pearl powder
分类算法 分类正确率/% 左右平移 叠加噪声 平移+噪声 DCGAN KNN 98.03 97.38 97.63 100 random forest 94.50 92.75 73.38 100 decision tree 95.87 98.87 86.63 100 1DCNN 99.78 99.12 99.04 100
Table 5. Comparison of quantitative models built by different data enhancement methods combined with 1DCNN
View tableView in ArticleTable 5. Comparison of quantitative models built by different data enhancement methods combined with 1DCNN
数据增强方法 R2 RMSEP LOSS 左右平移 0.856 2 0.125 4 0.015 6 叠加噪声 0.943 8 0.078 2 0.006 1 平移+噪声 0.844 0 0.130 3 0.017 0 DCGAN 0.988 4 0.034 8 0.001 2
Tools
Get Citation
Copy Citation Text
Ai-ling TAN, Zhen-yuan CHU, Xiao-si WANG, Yong ZHAO. Detection of Pearl Powder Adulteration Based on Raman Spectroscopy and DCGAN Data Enhancement[J]. Spectroscopy and Spectral Analysis, 2022, 42(3): 769
Paper Information
Category: Research Articles
Received: Feb. 7, 2021
Accepted: Apr. 19, 2021
Published Online: Apr. 19, 2022
The Author Email: TAN Ai-ling (tanailing@ysu.edu.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20