Quantitative Prediction of Heavy Metal Elements in white peony root Using Laser‐Induced Breakdown Spectroscopy and Semi‐Supervised Sequential Learning

Fudong Nian; Yujie Hu; Fuqiang Chen; Zhao Cheng; Yanhong Gu

doi:10.3788/CJL240790

Chinese Journal of Lasers, Volume. 51, Issue 23, 2311003(2024)

Quantitative Prediction of Heavy Metal Elements in white peony root Using Laser‐Induced Breakdown Spectroscopy and Semi‐Supervised Sequential Learning

Fudong Nian¹, Yujie Hu¹, Fuqiang Chen², Zhao Cheng³, and Yanhong Gu^1、*

Author Affiliations

¹Institute of Advanced Manufacturing Engineering, Hefei University, Hefei 230601, Anhui , China

²College of Environmental Science and Optoelectronic Technology, University of Science and Technology of China, Hefei 230026, Anhui , China

³Chinese Academy of Environmental Planning, Beijing 100043, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(14)

Fig. 1. Schematic diagram of LIBS experimental setup

Download full size

Fig. 2. Typical LIBS spectra of different white peony root samples. (a) Cd element; (b) Pb element

Download full size

Fig. 3. Semi-supervised sequence learning framework for predicting Pb and Cd contents based on white peony root LIBS data

Download full size

Fig. 4. Multilayer multichannel causal convolution model

Download full size

Fig. 5. Structure diagram of multiresolution one-dimensional sequential convolution model

Download full size

Fig. 6. Comparison of training loss values of two feature extraction modules with different optimizers. (a) DNN; (b) multiresolution one-dimensional sequential convolution

Download full size

Fig. 7. Training loss comparison for five deep learning-based feature extraction architectures

Download full size

Fig. 8. Test set fitting curves of five neural network feature extraction models

Download full size

Fig. 9. Training loss values comparison between fully supervised model using only labeled data and semi-supervised model using both labeled and unlabeled data

Download full size

Fig. 10. Quantitative prediction results of Pb and Cd elements in the test set using semi-supervised model using both labeled and unlabeled data and fully supervised model using only labeled data

Download full size

Fig. 11. Comparison of original and reconstructed LIBS spectra

Download full size

Table 1. Content of trace-heavy metal elements in white paeony root samples

View table

Table 1. Content of trace-heavy metal elements in white paeony root samples

Sample No.	Mass fraction /（mg/kg）		Sample No.	Mass fraction /（mg/kg）		Sample No.	Mass fraction /（mg/kg）
Sample No.	Pb	Cd	Sample No.	Pb	Cd	Sample No.	Pb	Cd
1#	17.35	14.92	20#	133.23	144.14	39#	75.69	85.52
2#	61.07	54.83	21#	134.63	188.20	40#	71.69	58.58
3#	79.66	97.37	22#	279.00	226.77	41#	32.56	38.95
4#	118.38	100.92	23#	231.62	173.31	42#	47.26	39.33
5#	142.20	188.61	24#	1888.02	1391.91	43#	31.46	17.40
6#	26.135	43.54	25#	195.32	177.22	44#	44.63	36.16
7#	221.94	368.25	26#	136.91	91.38	45#	15.58	19.02
8#	314.66	421.27	27#	268.98	252.69	46#	182.84	98.85
9#	182.65	120.59	28#	155.98	108.42	47#	325.44	305.74
10#	302.18	217.26	29#	730.20	519.34	48#	295.17	319.36
11#	281.81	208.83	30#	196.13	107.44	49#	260.74	209.52
12#	242.04	203.20	31#	220.22	174.59	50#	228.07	149.08
13#	125.95	85.17	32#	263.68	189.81	51#	136.32	100.88
14#	230.98	159.47	33#	443.14	287.69	52#	177.97	150.47
15#	160.15	111.98	34#	313.98	230.00	53#	298.72	207.42
16#	188.53	185.43	35#	233.05	167.59	54#	159.96	114.45
17#	104.42	88.52	36#	152.18	114.76	55#	286.63	247.93
18#	278.52	206.93	37#	486.27	357.05	56#	291.37	249.25
19#	118.17	97.55	38#	392.87	288.88	57#	255.54	206.64

Table 2. Average relative error of test set prediction for five neural network feature extraction models
View table
Table 2. Average relative error of test set prediction for five neural network feature extraction models
Model Average relative error /%
Pb Cd
Multiresolution onedimensional sequential convolution 5.54 5.16
DNN 4.11 31.22
LSTM 6.17 7.57
BiLSTM 6.82 6.97
Transformer 27.57 29.97

Table 3. Average relative error of predictions on test set by semi-supervised model using both labeled and unlabeled data and fully supervised model using only labeled data
View table
Table 3. Average relative error of predictions on test set by semi-supervised model using both labeled and unlabeled data and fully supervised model using only labeled data
Model Average relative error /%
Pb Cd
Fully supervised model using only labeled data 5.54 5.16
Semi-supervised model using both labeled and unlabeled data 4.12 3.32

Tools

Get Citation

Copy Citation Text

Fudong Nian, Yujie Hu, Fuqiang Chen, Zhao Cheng, Yanhong Gu. Quantitative Prediction of Heavy Metal Elements in white peony root Using Laser‐Induced Breakdown Spectroscopy and Semi‐Supervised Sequential Learning[J]. Chinese Journal of Lasers, 2024, 51(23): 2311003

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: spectroscopy

Received: Apr. 19, 2024

Accepted: May. 27, 2024

Published Online: Dec. 10, 2024

The Author Email: Gu Yanhong (guyh@hfuu.edu.cn)

DOI:10.3788/CJL240790

CSTR:32183.14.CJL240790

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Content of trace-heavy metal elements in white paeony root samples

Table 1. Content of trace-heavy metal elements in white paeony root samples

Table 2. Average relative error of test set prediction for five neural network feature extraction models

Table 2. Average relative error of test set prediction for five neural network feature extraction models

Table 3. Average relative error of predictions on test set by semi-supervised model using both labeled and unlabeled data and fully supervised model using only labeled data

Table 3. Average relative error of predictions on test set by semi-supervised model using both labeled and unlabeled data and fully supervised model using only labeled data