特征变量选择和回归方法相结合的土壤有机质含量估算

李冠稳; 高小红; 肖能文; 肖云飞

doi:doi:10.3788/AOS201939.0930002

光学学报, 2019, 39 (9): 0930002, 网络出版: 2019-09-09

特征变量选择和回归方法相结合的土壤有机质含量估算下载： 1176次

Estimation of Soil Organic Matter Content Based on Characteristic Variable Selection and Regression Methods

李冠稳 ^1,2,**高小红 ^1,*肖能文 ²肖云飞 ¹

作者单位

¹ 青海师范大学地理科学学院,青海省自然地理与环境过程重点实验室, 青海西宁 810008

² 中国环境科学研究院, 北京 100012

图 & 表

图 1. 研究区位置及采样点分布图

Fig. 1. Location of the study area and distribution of soil sampling sites

下载图片查看原文

图 2. 土壤样品反射率曲线。(a) 原始光谱;(b) MSC-MF-1st Derivative预处理光谱

Fig. 2. Spectral reflectance curves of soil samples. (a) Raw spectra; (b) spectra after MSC-MF-1st Derivative pre-processing

下载图片查看原文

图 3. sCARS变量筛选流程。(a)变量变化趋势;(b)十折交叉均方根误差;(c)变量回归系数

Fig. 3. Variable selection process by sCARS. (a) Changing trend of variables; (b) 10-fold RMSECV values; (c) regression coefficients of variables

下载图片查看原文

图 4. GA特征变量筛选过程

Fig. 4. Characteristic variable selection process by GA

下载图片查看原文

图 5. 预处理光谱sCARS-SPA特征变量筛选过程。(a)模型变量数;(b)变量指数

Fig. 5. Characteristic variable selection process by sCARS-SPA from the pre-processing spectrum. (a) Number of variables in the model; (b) variable index

下载图片查看原文

图 6. 不同变量筛选方法挑选特征变量分布

Fig. 6. Distribution of characteristic variables with different variable selection methods

下载图片查看原文

图 7. sCARS-PLSR模型预测SOM含量散点图

Fig. 7. Scatter plot for the measured and predicted value by sCARS-PLSR model

下载图片查看原文

图 8. SPA-SVM模型预测SOM含量散点图

Fig. 8. Scatter plot for the measured and predicted value by SPA-SVM model

下载图片查看原文

图 9. IRIV-RF模型预测SOM含量散点图

Fig. 9. Scatter plot for the measured and predicted value by IRIV-RF model

下载图片查看原文

图 10. 不同变量筛选方法PLSR、SVM、RF模型建模结果

Fig. 10. Results of PLSR, SVM and RF models with different variable selection methods

下载图片查看原文

图 11. 人工剔除异常值前后sCARS-RF模型散点图。(a)剔除异常值前;(b)剔除异常值后

Fig. 11. Scatter plots for the measured and predicted value by sCARS-RF model before and after artificially eliminating outliers. (a) Contain outliers; (b) eliminate outliers

下载图片查看原文

表 1校准集和验证集土壤有机质含量统计表

Table1. Soil organic matter content statistics of calibration set and validation set

Sample set	Samplenumber	Min /(g·kg^-1)	Max /(g·kg^-1)	Mean /(g·kg^-1)	SD
Calibration set	268	4.86	148.74	32.47	23.52
Validation set	133	8.26	133.56	32.16	22.44

查看原文

表 2不同变量筛选方法的PLSR模型精度

Table2. Accuracies of PLSR model with different variable selection methods

Selection method	Variable number	PC	Calibration set		Validation set
Selection method	Variable number	PC	$R_{cal}^{2}$	R_MSECAL	${R^{2}}_{val}$	R_MSEVAL	R_PD
Full-spectrum	2000	5	0.842	9.326	0.835	9.069	2.5
sCARS	51	5	0.874	8.327	0.883	7.797	2.9
SPA	5	5	0.850	9.103	0.858	8.525	2.6
GA	186	4	0.842	9.342	0.861	8.415	2.7
IRIV	63	6	0.843	9.300	0.875	8.043	2.8
sCARS-SPA	17	4	0.765	11.391	0.848	8.791	2.6

查看原文

表 3不同变量筛选方法的SVM建模精度

Table3. Accuracies of SVM model with different variable selection methods

Selection method	Variable number	Optimal parameter		Calibration set		Validation set
Selection method	Variable number	g (nuclear function)	c (punishment coefficient)	$R_{cal}^{2}$	R_MSECAL	$R_{val}^{2}$	R_MSEVAL	R_PD
Full-spectrum	2000	0.036	3.031	0.91	7.221	0.74	11.546	1.9
sCARS	51	0.021	3.031	0.881	8.116	0.877	7.918	2.8
SPA	5	0.004	1.741	0.858	8.855	0.889	7.477	3.0
GA	186	0.012	1.741	0.867	8.577	0.871	8.093	2.8
IRIV	63	0.021	3.031	0.869	8.493	0.864	8.307	2.7
sCARS-SPA	17	0.011	2.858	0.877	8.246	0.873	8.052	2.8

查看原文

表 4不同变量筛选方法RF建模精度

Table4. Accuracies of RF model with different variable selection methods

Selection method	Variable number	Calibration set		Validation set
Selection method	Variable number	$R_{cal}^{2}$	R_MSECAL	$R_{val}^{2}$	R_MSEVAL	R_PD
Full-spectrum	2000	0.942	5.817	0.957	4.840	4.6
sCARS	51	0.942	5.781	0.958	4.780	4.7
SPA	5	0.930	6.585	0.954	5.082	4.4
GA	186	0.939	5.894	0.959	4.699	4.8
IRIV	63	0.941	5.927	0.960	4.656	4.8
sCARS-SPA	17	0.940	5.910	0.955	4.971	4.5

查看原文

表 5人工剔除异常值后模型精度

Table5. Model accuracy after manually eliminating outliers

Model	Calibration set		Validation set
Model	$R_{cal}^{2}$	R_MSECAL	$R_{val}^{2}$	R_MSEVAL	R_PD
sCARS-PLSR	0.943	5.538	0.926	5.987	3.7
sCARS-SVM	0.926	6.092	0.957	4.903	4.6
sCARS-RF	0.985	3.204	0.988	2.865	7.8

查看原文

李冠稳, 高小红, 肖能文, 肖云飞. 特征变量选择和回归方法相结合的土壤有机质含量估算[J]. 光学学报, 2019, 39(9): 0930002. Guanwen Li, Xiaohong Gao, Nengwen Xiao, Yunfei Xiao. Estimation of Soil Organic Matter Content Based on Characteristic Variable Selection and Regression Methods[J]. Acta Optica Sinica, 2019, 39(9): 0930002.

特征变量选择和回归方法相结合的土壤有机质含量估算下载： 1176次

图 1. 研究区位置及采样点分布图

Fig. 1. Location of the study area and distribution of soil sampling sites

图 2. 土壤样品反射率曲线。(a) 原始光谱;(b) MSC-MF-1st Derivative预处理光谱

Fig. 2. Spectral reflectance curves of soil samples. (a) Raw spectra; (b) spectra after MSC-MF-1st Derivative pre-processing

图 3. sCARS变量筛选流程。(a)变量变化趋势;(b)十折交叉均方根误差;(c)变量回归系数

Fig. 3. Variable selection process by sCARS. (a) Changing trend of variables; (b) 10-fold RMSECV values; (c) regression coefficients of variables

图 4. GA特征变量筛选过程

Fig. 4. Characteristic variable selection process by GA

图 5. 预处理光谱sCARS-SPA特征变量筛选过程。(a)模型变量数;(b)变量指数

Fig. 5. Characteristic variable selection process by sCARS-SPA from the pre-processing spectrum. (a) Number of variables in the model; (b) variable index

图 6. 不同变量筛选方法挑选特征变量分布

Fig. 6. Distribution of characteristic variables with different variable selection methods

图 7. sCARS-PLSR模型预测SOM含量散点图

Fig. 7. Scatter plot for the measured and predicted value by sCARS-PLSR model

图 8. SPA-SVM模型预测SOM含量散点图

Fig. 8. Scatter plot for the measured and predicted value by SPA-SVM model

图 9. IRIV-RF模型预测SOM含量散点图

Fig. 9. Scatter plot for the measured and predicted value by IRIV-RF model

图 10. 不同变量筛选方法PLSR、SVM、RF模型建模结果

Fig. 10. Results of PLSR, SVM and RF models with different variable selection methods

图 11. 人工剔除异常值前后sCARS-RF模型散点图。(a)剔除异常值前;(b)剔除异常值后

Fig. 11. Scatter plots for the measured and predicted value by sCARS-RF model before and after artificially eliminating outliers. (a) Contain outliers; (b) eliminate outliers

表 1校准集和验证集土壤有机质含量统计表

Table1. Soil organic matter content statistics of calibration set and validation set

表 2不同变量筛选方法的PLSR模型精度

Table2. Accuracies of PLSR model with different variable selection methods

表 3不同变量筛选方法的SVM建模精度

Table3. Accuracies of SVM model with different variable selection methods

表 4不同变量筛选方法RF建模精度

Table4. Accuracies of RF model with different variable selection methods

表 5人工剔除异常值后模型精度

Table5. Model accuracy after manually eliminating outliers

关于本站 Cookie 的使用提示

全站搜索

特征变量选择和回归方法相结合的土壤有机质含量估算 下载： 1176次

图 1. 研究区位置及采样点分布图

Fig. 1. Location of the study area and distribution of soil sampling sites

图 2. 土壤样品反射率曲线。(a) 原始光谱;(b) MSC-MF-1st Derivative预处理光谱

Fig. 2. Spectral reflectance curves of soil samples. (a) Raw spectra; (b) spectra after MSC-MF-1st Derivative pre-processing

图 3. sCARS变量筛选流程。(a)变量变化趋势;(b)十折交叉均方根误差;(c)变量回归系数

Fig. 3. Variable selection process by sCARS. (a) Changing trend of variables; (b) 10-fold RMSECV values; (c) regression coefficients of variables

图 4. GA特征变量筛选过程

Fig. 4. Characteristic variable selection process by GA

图 5. 预处理光谱sCARS-SPA特征变量筛选过程。(a)模型变量数;(b)变量指数

Fig. 5. Characteristic variable selection process by sCARS-SPA from the pre-processing spectrum. (a) Number of variables in the model; (b) variable index

图 6. 不同变量筛选方法挑选特征变量分布

Fig. 6. Distribution of characteristic variables with different variable selection methods

图 7. sCARS-PLSR模型预测SOM含量散点图

Fig. 7. Scatter plot for the measured and predicted value by sCARS-PLSR model

图 8. SPA-SVM模型预测SOM含量散点图

Fig. 8. Scatter plot for the measured and predicted value by SPA-SVM model

图 9. IRIV-RF模型预测SOM含量散点图

Fig. 9. Scatter plot for the measured and predicted value by IRIV-RF model

图 10. 不同变量筛选方法PLSR、SVM、RF模型建模结果

Fig. 10. Results of PLSR, SVM and RF models with different variable selection methods

图 11. 人工剔除异常值前后sCARS-RF模型散点图。(a)剔除异常值前;(b)剔除异常值后

Fig. 11. Scatter plots for the measured and predicted value by sCARS-RF model before and after artificially eliminating outliers. (a) Contain outliers; (b) eliminate outliers

表 1校准集和验证集土壤有机质含量统计表

Table1. Soil organic matter content statistics of calibration set and validation set

表 2不同变量筛选方法的PLSR模型精度

Table2. Accuracies of PLSR model with different variable selection methods

表 3不同变量筛选方法的SVM建模精度

Table3. Accuracies of SVM model with different variable selection methods

表 4不同变量筛选方法RF建模精度

Table4. Accuracies of RF model with different variable selection methods

表 5人工剔除异常值后模型精度

Table5. Model accuracy after manually eliminating outliers

相关论文

相关资讯

关于本站 Cookie 的使用提示

全站搜索

特征变量选择和回归方法相结合的土壤有机质含量估算下载： 1176次