Regression Prediction of Soil Available Nitrogen Near-Infrared Spectroscopy Based on Boosting Algorithm

Yalu Han; Shaowen Li; Wenrui Zheng; Shengqun Shi; Xianzhi Zhu; Xiu Jin

doi:10.3788/LOP202158.1630005

Journals >Laser & Optoelectronics Progress >Volume 58 >Issue 16 >Page 1630005 > Article

Laser & Optoelectronics Progress
Vol. 58, Issue 16, 1630005 (2021)

Regression Prediction of Soil Available Nitrogen Near-Infrared Spectroscopy Based on Boosting Algorithm

Yalu Han, Shaowen Li^*, Wenrui Zheng, Shengqun Shi, Xianzhi Zhu, and Xiu Jin

Author Affiliations

School of Information & Computer, Anhui Agricultural University, Hefei, Anhui 230036, China

show less

DOI: 10.3788/LOP202158.1630005 Cite this Article Set citation alerts

Yalu Han, Shaowen Li, Wenrui Zheng, Shengqun Shi, Xianzhi Zhu, Xiu Jin. Regression Prediction of Soil Available Nitrogen Near-Infrared Spectroscopy Based on Boosting Algorithm[J]. Laser & Optoelectronics Progress, 2021, 58(16): 1630005 Copy Citation Text

show less

Fig. 1. Indoor spectrum acquisition system

Download full size

Fig. 2. Boosting algorithm classification

Download full size

Fig. 3. Schematic of AdaBoost regression

Download full size

Fig. 4. Technical route analysis of near-infrared hyperspectral characteristics of available nitrogen in soil

Download full size

Fig. 5. Contrast of spectra before and after preprocess. (a) Original spectra; (b) SG; (c) LG; (d) FD; (e) SNV; (f) SG+SNV; (g) SG+LG; (h) SG+FD

Download full size

Fig. 6. R² and RPD values of regression models with testing obtained by different pretreatment methods. (a) R²; (b) RPD

Download full size

Fig. 7. R², RMSE, and RPD values of the testing sets of different algorithms. (a) R²;(b) RMSE; (c) RPD

Download full size

Fig. 8. Optimal combination of wavelength points selected by different algorithms

Download full size

Fig. 9. Measured and predicted values of PSO-AdaBoost model based on SNV in prediction set

Download full size

Dataset	Number of samples	Max /(mg·kg^-1)	Min /(mg·kg^-1)	Median /(mg·kg^-1)	Mean /(mg·kg^-1)	Standard /(mg·kg^-1)
Total set	188	731.584	19.32	132.644	179.083	144.398
Training set	131	731.584	19.32	130.732	179.440	148.217
Testing set	57	687.148	67.62	137.816	178.255	135.211

Table 1. Statistical parameters of soil available nitrogen content

Preprocess method	Regression model	Training set		Testing set		Parameter
Preprocess method	Regression model	R²	RPD	R²	RPD	Number of latent variables
SG	PLSR	0.92	3.62	0.894	3.08	16
LG	PLSR	0.94	4.02	0.898	3.14	11
FD	PLSR	0.95	4.39	0.718	1.88	6
SNV	PLSR	0.96	4.87	0.857	2.64	11
SG+SNV	PLSR	0.94	4.16	0.845	2.54	11
SG+LG	PLSR	0.95	4.25	0.916	3.45	14
SG+FD	PLSR	0.88	2.87	0.738	1.95	6
Preprocess method	Regression model	Training set		Testing set		Parameter
Preprocess method	Regression model	R²	RPD	R²	RPD	Learning rate/number of estimators
SG	GBRT	0.99	28.27	0.610	1.60	0.4/400
LG	GBRT	0.99	28.27	0.508	1.43	0.2/200
FD	GBRT	0.99	28.27	0.668	1.73	0.4/200
Preprocess method	Regression model	Training set		Testing set		Parameter
Preprocess method	Regression model	R²	RPD	R²	RPD	Learning rate/number of estimators
SNV	GBRT	0.99	34.96	0.915	3.43	0.2/100
SG+SNV	GBRT	0.99	15.13	0.910	3.33	0.4/100
SG+LG	GBRT	0.99	28.27	0.573	1.53	0.4/100
SG+FD	GBRT	0.99	22.27	0.898	3.14	0.1/300
SG	AdaBoost	0.97	5.43	0.644	1.68	0.4/100
LG	AdaBoost	0.95	4.68	0.576	1.54	0.4/200
FD	AdaBoost	0.99	23.74	0.573	1.53	0.1/200
SNV	AdaBoost	0.99	12.14	0.921	3.43	0.2/100
SG+SNV	AdaBoost	0.99	20.07	0.912	3.37	0.1/200
SG+LG	AdaBoost	0.96	4.75	0.319	1.21	0.3/200
SG+FD	AdaBoost	0.99	28.27	0.876	2.84	0.1/200
SG	XGBoost	0.99	28.24	0.745	1.98	0.1/300
LG	XGBoost	0.99	28.24	0.739	1.95	0.1/300
FD	XGBoost	0.99	28.26	0.470	1.15	0.2/100
SNV	XGBoost	0.99	28.21	0.912	3.37	0.4/100
SG+SNV	XGBoost	0.99	25.26	0.908	3.31	0.2/100
SG+LG	XGBoost	0.99	28.25	0.745	1.98	0.1/300
SG+FD	XGBoost	0.99	28.26	0.835	2.46	0.4/100
SG	LightGBM	0.79	2.21		0.81	0.1/400
LG	LightGBM	0.75	1.99		0.69	0.4/400
FD	LightGBM	0.99	19.60	0.521	1.44	0.4/200
SNV	LightGBM	0.99	26.87	0.849	2.57	0.4/100
SG+SNV	LightGBM	0.99	25.44	0.857	2.65	0.4/100
SG+LG	LightGBM	0.79	2.19		0.68	0.3/400
SG+FD	LightGBM	0.99	22.62	0.695	1.81	0.1/200

Table 2. Influence of different pretreatment methods on regression model

Preprocess method		Algorithm	Wavelength range /nm	Number of variables	Training set		Testing set		Parameter
Preprocess method		Algorithm	Wavelength range /nm	Number of variables	R²	RPD	R²	RPD	Learning rate/number of estimators
SNV	AdaBoost		350--1655	1305	0.99	12.14	0.921	3.43	0.2/100
SNV	GBRT		350--1655	1305	0.99	34.96	0.915	3.43	0.2/100
SNV	XGBoost		350--1655	1305	0.99	28.21	0.912	3.37	0.4/100
SNV	RF-GBRT		600--999	200	0.99	28.27	0.922	3.57	0.4/300
SNV	PSO-GBRT		602--999	202	0.99	28.27	0.924	3.63	0.2/400
SNV	GA-GBRT		600--999	184	0.99	28.27	0.932	3.83	0.4/300
SNV	SA-GBRT		602--998	206	0.99	28.27	0.941	4.11	0.4/300
SNV	GGA-GBRT		601--999	202	0.99	28.27	0.919	3.52	0.2/400
SNV	RF-AdaBoost		600--999	200	0.99	21.24	0.939	4.06	0.2/100
SNV	PSO-AdaBoost		602--999	202	0.99	18.17	0.944	4.24	0.1/100
SNV	GA- AdaBoost		600--999	184	0.99	12.95	0.940	4.09	0.4/100
SNV	SA-AdaBoost		602--998	206	0.99	24.03	0.937	3.96	0.2/100
SNV	GGA-AdaBoost		601--999	202	0.99	24.21	0.943	4.20	0.4/100
SNV	RF-XGBoost		600--999	200	0.99	28.17	0.929	3.76	0.2/100
SNV	PSO-XGBoost		602--999	202	0.99	28.25	0.821	3.36	0.2/400
SNV	GA-XGBoost		600--999	184	0.99	28.22	0.886	2.96	0.3/100
SNV	SA-XGBoost		602--998	206	0.99	20.21	0.834	2.46	0.1/200
SNV	GGA-XGBoost		601--999	202	0.99	27.96	0.871	2.78	0.1/200
SG+SNV	AdaBoost		350--1655	1305	0.99	20.07	0.912	3.37	0.1/200
SG+SNV	GBRT		350--1655	1305	0.99	15.13	0.910	3.33	0.4/100
SG+SNV	XGBoost		350--1655	1305	0.99	25.26	0.908	3.31	0.2/100
SG+SNV	RF-GBRT		603--999	202	0.99	28.27	0.919	3.53	0.2/200
SG+SNV	PSO-GBRT		607--999	201	0.99	28.27	0.913	3.39	0.1/300
SG+SNV	GA-GBRT		604--999	209	0.99	28.27	0.927	3.69	0.4/100
Preprocess method		Algorithm	Wavelength range /nm	Number of variables	Training set		Testing set		Parameter
Preprocess method		Algorithm	Wavelength range /nm	Number of variables	R²	RPD	R²	RPD	Learning rate/number of estimators
SG+SNV	SA-GBRT		607--998	206	0.99	28.27	0.926	3.68	0.4/200
SG+SNV	GGA-GBRT		607--999	200	0.99	28.27	0.900	3.17	0.3/400
SG+SNV	RF-AdaBoost		603--999	202	0.99	12.12	0.919	3.51	0.3/100
SG+SNV	PSO-AdaBoost		607--999	201	0.99	15.41	0.915	3.43	0.4/100
SG+SNV	GA-AdaBoost		604--999	209	0.99	24.16	0.922	3.59	0.2/100
SG+SNV	SA-AdaBoost		607--998	206	0.99	20.24	0.929	3.76	0.1/100
SG+SNV	GGA-AdaBoost		607--999	200	0.99	12.45	0.924	3.64	0.3/300
SG+SNV	RF-XGBoost		600--999	200	0.99	28.18	0.898	3.14	0.4/200
SG+SNV	PSO-XGBoost		602--999	202	0.99	27.45	0.888	2.99	0.4/100
SG+SNV	GA-XGBoost		600--999	184	0.99	24.22	0.885	2.95	0.3/100
SG+SNV	SA-XGBoost		602--998	206	0.99	27.31	0.883	2.92	0.4/100
SG+SNV	GGA-XGBoost		601--999	202	0.99	24.84	0.882	2.91	0.3/100

Table 3. Analysis results of quantitative models with different variables based on the spectral data processed by SNV and SG+SNV

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information