Author Affiliations
1College of Geology Engineering and Surveying, Chang’an University, Xi’an 710054, Shaanxi , China2Key laboratory of Degraded and Unused Land Consolidation Engineering, Ministry of Natural Resources, Xi’an 710016, Shaanx , China3Technologies Co., Ltd., Xi’an 710001, Shaanxi , China4Changqing Engineering Design Co., Ltd., Xi’an 710018, Shaanxi , Chinashow less
Fig. 1. Mean-standard deviation distribution of MCCV method
Fig. 2. Spectral curves of different kinds of spectral transformations. (a) R; (b) CR; (c) log R-1
Fig. 3. Correlation analysis of different spectral transformations and heavy metal content
Fig. 4. Scatter diagrams of prediction results of CFS-XGBoost. (a) R; (b) log R-1; (c) CR
Fig. 5. Scatter diagram of prediction results of GA-XGBoost. (a) R; (b) log R-1; (c) CR
Fig. 6. Results of correlation coefficient feature selection and GA feature selection
Fig. 7. Feature importance scores given by XGBoost
Fig. 8. Spatial distribution of Cu content
Satellite and its sensor | Band | Wavelength /nm | Spectral resolution /nm | Spatial resolution /m |
---|
GF-5(AHSI) | 330 | 450‒2500 | 5(VNIR),10(SWIR) | 30 | HJ-1-A(HSI) | 105 | 450‒1050 | 2‒9 | 100 | EO-1(Hyperion) | 220 | 400‒2500 | 10 | 30 |
|
Table 1. Comparison of hyperspectral satellite parameters
Item | Minimum | Median | Maximum | Mean |
---|
Background | 6.8 | 19.5 | 43.6 | 21.4 | Sample | 24.0 | 46.0 | 75.0 | 48.2 |
|
Table 2. Comparison between measured Cu content and regional background value
Spectrum transform | Maximum absolute correlation coefficient | Band | Number of significant bands |
---|
R | 0.464** | R2344 | 77 | CR | 0.590** | R2344 | 95 | log R-1 | 0.453** | R2344 | 186 |
|
Table 3. Statistics of correlation coefficient between original spectrum and its two transformations with Cu content
Sample set | Sample size | Maximum /(mg·kg-1) | Minimum /(mg·kg-1) | Mean /(mg·kg-1) | Standard deviation /(mg·kg-1) |
---|
Total set | 39 | 75.0 | 24.0 | 48.2 | 13.38 | Training set | 31 | 75.0 | 26.0 | 49.1 | 13.5 | Testing set | 8 | 64.0 | 24.0 | 45.0 | 13.1 |
|
Table 4. Statistical characteristics of training set and testing set
method Selection | Model | Training set | Testing set | Number of bands |
---|
R2 | RMSE | RPD | R2 | RMSE | RPD |
---|
CFS | R-CFS-XGB | 0.95 | 3.05 | 4.1 | 0.51 | 8.57 | 1.1 | 77 | CR-CFS-XGB | 0.92 | 3.77 | 3.0 | 0.61 | 7.59 | 1.6 | 95 | log R-1-CFS-XGB | 0.94 | 3.12 | 3.9 | 0.53 | 8.36 | 1.3 | 186 | GA | R-GA-XGB | 0.95 | 3.07 | 3.8 | 0.61 | 7.60 | 1.5 | 139 | CR-GA-XGB | 0.94 | 3.38 | 3.3 | 0.84 | 4.85 | 2.0 | 137 | log R-1-GA-XGB | 0.91 | 3.86 | 2.7 | 0.65 | 7.48 | 1.7 | 110 |
|
Table 5. Precision comparison of XGBoost model based on GA and CFS feature selection methods
Content of copper /(mg·kg-1) | 19.79‒21.40 | 21.40‒38.76 | 38.76‒45.02 | 45.02‒51.83 | 51.83‒66.75 |
---|
Percentage /% | 0.07 | 53.06 | 21.07 | 17.86 | 7.94 |
|
Table 6. Statistics of copper content estimation results