Application of Various Algorithms for Spectral Variable Selection in NIRS Modeling of Red Ginseng Extraction

Bei CHEN; En-rang ZHENG; Tuo GUO

doi:10.3964/j.issn.1000-0593(2021)08-2443-07

Journals >Spectroscopy and Spectral Analysis >Volume 41 >Issue 8 >Page 2443 > Article

Spectroscopy and Spectral Analysis
Vol. 41, Issue 8, 2443 (2021)

Application of Various Algorithms for Spectral Variable Selection in NIRS Modeling of Red Ginseng Extraction

Bei CHEN^1、*, En-rang ZHENG^{1、1; *;}, and Tuo GUO^2、2;

Author Affiliations

¹1. School of Electrical and Control Engineering, Shaanxi University of Science & Technology, Xi’an 710021, China

²2. School of Electronic Information and Artificial Intelligence, Shaanxi University of Science & Technology, Xi’an 710021, China

show less

DOI: 10.3964/j.issn.1000-0593(2021)08-2443-07 Cite this Article

Bei CHEN, En-rang ZHENG, Tuo GUO. Application of Various Algorithms for Spectral Variable Selection in NIRS Modeling of Red Ginseng Extraction[J]. Spectroscopy and Spectral Analysis, 2021, 41(8): 2443 Copy Citation Text

show less

Fig. 1. Flow chart of NIRS modeling process of red ginseng extraction

Download full size | View in the Article

Fig. 2. NIR spectra of 128 red ginseng samples

Download full size | View in the Article

Fig. 3. Comparison of optimal detection wavelength distribution of ginsenoside Rg1 content

Download full size | View in the Article

Fig. 4. Distribution diagrams of ginsenoside Rg1 content predicted by different modeling methods

Download full size | View in the Article

项目	统计量	指标含量/(mg·mL^-1)
项目	统计量	Rg1	Rc
校正集93个	最小值	0.127 5	0.057 7
	最大值	0.506 2	0.278 3
	平均值	0.284 3	0.158 2
	标准差	0.120 3	0.069 9
验证集31个	最小值	0.124 0	0.071 4
	最大值	0.502 7	0.286 6
	平均值	0.258 6	0.142 4
	标准偏差	0.123 3	0.071 6

Table 1. Statistical table of sample division

View in the Article

成分	方法	主成分数	特征波长数	校正集		验证集
成分	方法	主成分数	特征波长数	$R_{C}^{2}$	RMSECV	$R_{P}^{2}$	RMSEP	RPD
Rg1	Full-PLS	5	1 100	0.898 1	0.037 7	0.920 2	0.033 7	3.540 7
	CARS-PLS	4	57	0.942 1	0.028 4	0.893 0	0.039 0	3.057 0
	UVE-PLS	5	430	0.917 9	0.033 8	0.912 8	0.035 2	3.386 2
	SPA-PLS	5	8	0.849 2	0.045 9	0.915 7	0.034 6	3.445 0
	RF-PLS	4	10	0.946 4	0.027 4	0.957 4	0.024 6	4.845 3
Rc	Full-PLS	5	1 100	0.927 7	0.018 2	0.920 7	0.020 4	3.551 3
	CARS-PLS	4	65	0.960 5	0.013 5	0.917 5	0.020 8	3.481 1
	UVE-PLS	5	131	0.916 5	0.019 6	0.920 7	0.020 4	3.562 0
	SPA-PLS	5	5	0.859 6	0.025 4	0.944 1	0.017 1	4.228 1
	RF-PLS	3	10	0.963 1	0.013 0	0.965 1	0.013 5	5.351 5

Table 2. Different characteristic wavelength variable selection and performance evaluation results of red ginsenoside content based on PLS model

View in the Article

成分	方法	主成分数	特征波长数		校正集				验证集
成分	方法	主成分数	特征波长数		$R_{C}^{2}$		RMSECV		$R_{P}^{2}$		RMSEP		RPD
Rg1	原始光谱RF	5	10	0.921 4		0.033 5		0.931 6		0.031 7		3.822 5
	全光谱RF	5	10	0.944 4		0.028 0		0.948 4		0.027 0		4.403 2
	全光谱SNV+RF	4	10	0.946 4		0.027 4		0.957 4		0.024 6		4.845 3
Rc	原始光谱RF	5	10	0.957 7		0.014 3		0.958 9		0.014 3		4.933 8
	全光谱RF	5	10	0.970 7		0.011 5		0.960 3		0.014 7		5.018 9
	全光谱SNV+RF	3	10	0.963 1		0.013 0		0.965 1		0.013 5		5.351 5

Table 3. PLS modeling and performance evaluation of different spectra based on RF algorithm

View in the Article

成分	方法	主成分数	特征波长数		校正集				验证集
成分	方法	主成分数	特征波长数		$R_{C}^{2}$		RMSECV		$R_{P}^{2}$		RMSEP		RPD
Rg1	原始光谱CARS	5	59	0.703 7		0.065 1		-0.343 0		0.140 6		0.862 9
	全光谱CARS	5	9	0.908 4		0.036 0		0.857 0		0.044 9		2.644 2
	全光谱SNV+CARS	4	57	0.942 1		0.028 4		0.893 0		0.039 0		3.057 0
Rc	原始光谱CARS	5	51	0.743 2		0.035 2		-0.243 2		0.078 5		0.896 7
	全光谱CARS	5	65	0.952 2		0.014 7		0.908 8		0.022 3		3.312 2
	全光谱SNV+CARS	4	65	0.960 5		0.013 5		0.917 5		0.020 8		3.481 1

Table 4. PLS modeling and performance evaluation of different spectra based on CARS algorithm

Download Citation

Save the article for my favorites

Paper Information