Hyperspectral Inversion of Soil Selenium Content Based on Seagull Algorithm Optimized Random Forest

Peng Xie; Zhenghai Wang; Bei Xiao; Yuxin Tian

doi:10.3788/LOP222037

Laser & Optoelectronics Progress, Volume. 60, Issue 17, 1730001(2023)

Hyperspectral Inversion of Soil Selenium Content Based on Seagull Algorithm Optimized Random Forest

Peng Xie, Zhenghai Wang^*, Bei Xiao, and Yuxin Tian

Author Affiliations

School of Earth Sciences and Engineering, Sun Yat-sen University, Guangzhou 510275, Guangdong , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(15)

Fig. 1. Technology roadmap

Download full size

Fig. 2. Spectral curves of soil samples. (a) Original spectra; (b) MSC-FD transformed spectra

Download full size

Fig. 3. SOA-RF Flowchart

Download full size

Fig. 4. Correlation analysis. (a) Original spectra; (b) MSC-FD

Download full size

Fig. 5. sCARS algorithm feature band screening process

Download full size

Fig. 6. VISSA algorithm feature band screening process. (a) RMSECV values; (b) weight values

Download full size

Fig. 7. Distribution of wavelength of features extracted by different algorithms

Download full size

Fig. 8. SOA-RF algorithm convergence curve

Download full size

Fig. 9. SOA-RF model prediction results(RPD)

Download full size

Fig. 10. Scatter plot of measured and predicted values of the optimal model. (a) sCARS-PLSR; (b) IRIV-SVM; (c) sCARS-RF; (d) VCPA-GA-SOA-RF

Download full size

Table 1. Descriptive statistics of soil selenium content in training and prediction sets
View table
Table 1. Descriptive statistics of soil selenium content in training and prediction sets
Sample Number of sample Minimum value /（μg·g^-1） Maximum value /（μg·g^-1） Average /（μg·g^-1） Standard deviation
Training set 35 0.200 1.150 0.726 0.256
Prediction set 14 0.295 1.025 0.595 0.213

Table 2. Evaluation of accuracy of the PLSR model with different variable screenings

View table

Table 2. Evaluation of accuracy of the PLSR model with different variable screenings

Spectral transform	Time /s	Number of variable	Training R²	Training RMSE	Validation R²	Validation RMSE
Full spectra	‒	2051	0.43	0.21	0.37	0.35
sCARS	<10.00	10	0.52	0.17	0.49	0.19
CARS	<10.00	14	0.52	0.19	0.40	0.16
IRIV	145.60	13	0.65	0.15	0.49	0.18
VISSA	51.96	65	0.49	0.19	0.48	0.16
IVISSA	27.11	73	0.50	0.15	0.45	0.20
VCPA-GA	145.53	42	0.54	0.15	0.51	0.19

Table 3. Evaluation of accuracy of SVM models with different variable screenings

View table

Table 3. Evaluation of accuracy of SVM models with different variable screenings

Spectral transform	Time /s	Number of variable	Training R²	Training RMSE	Validation R²	Validation RMSE
Full spectra	‒	2051	0.57	0.05	0.34	0.66
sCARS	<10.00	10	0.59	0.11	0.57	0.14
CARS	<10.00	14	0.58	0.13	0.55	0.12
IRIV	145.60	13	0.66	0.13	0.55	0.11
VISSA	51.96	65	0.77	0.13	0.53	0.13
IVISSA	27.11	73	0.96	0.01	0.56	0.39
VCPA-GA	145.53	42	0.56	0.11	0.54	0.14

Table 4. Evaluation of accuracy of the RF model with different variable screenings

View table

Table 4. Evaluation of accuracy of the RF model with different variable screenings

Spectral transform	Time /s	Number of variable	Training R²	Training RMSE	Validation R²	Validation RMSE
Full spectra	‒	2051	0.85	0.19	0.47	0.18
sCARS	<10.00	10	0.82	0.09	0.66	0.13
CARS	<10.00	14	0.62	0.16	0.49	0.17
IRIV	145.60	13	0.86	0.06	0.68	0.15
VISSA	51.96	65	0.86	0.05	0.65	0.16
IVISSA	27.11	73	0.81	0.06	0.64	0.15
VCPA-GA	145.53	42	0.69	0.11	0.59	0.19

Table 5. Evaluation of accuracy of the SOA-RF model with different variable screenings

View table

Table 5. Evaluation of accuracy of the SOA-RF model with different variable screenings

Spectral transform	Time /s	Number of variable	Training R²	Training RMSE	Validation R²	Validation RMSE
Full spectra	‒	2051	0.85	0.16	0.75	0.15
sCARS	<10.00	10	0.93	0.07	0.90	0.06
CARS	<10.00	14	0.91	0.08	0.80	0.11
IRIV	145.60	13	0.86	0.10	0.81	0.11
VISSA	51.96	65	0.94	0.07	0.84	0.10
IVISSA	27.11	73	0.94	0.06	0.83	0.09
VCPA-GA	145.53	42	0.94	0.07	0.92	0.08

Tools

Get Citation

Copy Citation Text

Peng Xie, Zhenghai Wang, Bei Xiao, Yuxin Tian. Hyperspectral Inversion of Soil Selenium Content Based on Seagull Algorithm Optimized Random Forest[J]. Laser & Optoelectronics Progress, 2023, 60(17): 1730001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Spectroscopy

Received: Jul. 11, 2022

Accepted: Aug. 29, 2022

Published Online: Sep. 13, 2023

The Author Email: Zhenghai Wang (wzhengh@mail.sysu.edu.cn)

DOI:10.3788/LOP222037

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Descriptive statistics of soil selenium content in training and prediction sets

Table 1. Descriptive statistics of soil selenium content in training and prediction sets

Table 2. Evaluation of accuracy of the PLSR model with different variable screenings

Table 2. Evaluation of accuracy of the PLSR model with different variable screenings

Table 3. Evaluation of accuracy of SVM models with different variable screenings

Table 3. Evaluation of accuracy of SVM models with different variable screenings

Table 4. Evaluation of accuracy of the RF model with different variable screenings

Table 4. Evaluation of accuracy of the RF model with different variable screenings

Table 5. Evaluation of accuracy of the SOA-RF model with different variable screenings

Table 5. Evaluation of accuracy of the SOA-RF model with different variable screenings