Chinese Optics Letters, Volume. 21, Issue 4, 043001(2023)
Rapid classification of copper concentrate by portable laser-induced breakdown spectroscopy combined with transfer learning and deep convolutional neural network
Figures & Tables(14)
Fig. 2. Typical spectrum of copper concentrate acquired by portable LIBS apparatus.
Fig. 3. Elemental spectral line intensities of the raw spectra of copper concentrates from 11 classes.
Fig. 4. PCA plot of the raw spectra of copper concentrates from 11 classes.
Fig. 6. Schematic diagram of 2D spectrum (left) and 2D spectrum image (right).
Fig. 8. Confusion matrices of the four CNN models on the test set (two upper panels, VGG16 and ResNet18; two lower panels, DenseNet121 and InceptionV3).
Fig. 9. Comparison of classification accuracy between CNN models and traditional machine-learning models on the test set.
Fig. 10. Schematic diagram of the spectral features selected by the CST methods.
Table 1. Copper Contents of the Copper Concentrate Samples
View tableView in ArticleTable 1. Copper Contents of the Copper Concentrate Samples
No. 1 2 3 4 5 6 7 8 9 10 11 Mass fraction (%) 18.86 19.68 20.58 21.80 22.49 23.66 24.42 25.52 26.21 27.73 27.62
Table 2. Performance of the CNN Models with the Convolutional Layers Frozen and Only Fully Connected Layers Trained
View tableView in ArticleTable 2. Performance of the CNN Models with the Convolutional Layers Frozen and Only Fully Connected Layers Trained
Model Training Accuracy Validation Accuracy Test Accuracy VGG16 73.4% 82.2% 79.9% ResNet18 71.8% 68.6% 69.3% DenseNet121 76.4% 59.1% 51.1% InceptionV3 70.7% 65.9% 65.5%
Table 3. Performance of the CNN Model with the Convolutional Layer Unfrozen and Trained with All Parameters
View tableView in ArticleTable 3. Performance of the CNN Model with the Convolutional Layer Unfrozen and Trained with All Parameters
Model Training Accuracy Validation Accuracy Test Accuracy VGG16 99.7% 95.1% 96.2% ResNet18 100% 94.7% 92.8% DenseNet121 100% 94.3% 93.6% InceptionV3 100% 93.6% 93.6% NPT-VGG a 99.1% 88.63% 89.4% NPT-ResNet 100% 85.2% 83.3% NPT-DenseNet 100% 62.1% 61.4% NPT-Inception 100% 80.7% 81.1%
Table 4. Performance of the Four Machine-Learning Models
View tableView in ArticleTable 4. Performance of the Four Machine-Learning Models
Model Training Accuracy Validation Accuracy Test Accuracy PCA-BPNN 100% 92.5% 91.3% PCA-SVM 98.7% 89.8% 85.2% CST-BPNN 100% 92.4% 90.5% CST-SVM 100% 88.3% 87.5%
Tools
Get Citation
Copy Citation Text
Haochen Li, Tianyuan Liu, Yuchao Fu, Wanxiang Li, Meng Zhang, Xi Yang, Di Song, Jiaqi Wang, You Wang, Meizhen Huang. Rapid classification of copper concentrate by portable laser-induced breakdown spectroscopy combined with transfer learning and deep convolutional neural network[J]. Chinese Optics Letters, 2023, 21(4): 043001
Paper Information
Category: Spectroscopy
Received: Oct. 13, 2022
Accepted: Nov. 28, 2022
Published Online: Apr. 10, 2023
The Author Email: Tianyuan Liu (tianyuanl@sjtu.edu.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20