Fig. 1. Near Infrared microplastic measuring system
1: Optical fiber; 2: Sample; 3: Integrating sphere with light source; 4: Receiving optical fiber; 5: Spectral collection module; 6: Spectral processing module; 7: Photoelectric conversion module; 8: Date processing module
Fig. 2. K-fold cross-validation structure
Fig. 3. XGBoost and SVM for microplastic detection
Fig. 4. Original Near Infrared spectra of microplastics
Fig. 5. The 30 most important spectral variables
Fig. 6. CV average accuracy of n_estimators, learning_rate, max_depth, gamma
Fig. 7. SVM and XGBoost confusion matrixes
Fig. 8. Accuracy score, precision score, recall and F1-score of 10-fold cross validations of XGBoost and SVM
Microplastic category | Precision of XGBoost | Recall of XGBoost | F1-score of XGBoost | Precision of SVM | Recall of SVM | F1-score of SVM |
---|
ABS | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 | EVA | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 | PAN | 0.93 | 1.00 | 0.97 | 0.82 | 1.00 | 0.90 | PBT | 1.00 | 1.00 | 1.00 | 1.00 | 0.84 | 0.91 | PC | 0.86 | 1.00 | 0.92 | 1.00 | 0.89 | 0.94 | PCL | 0.91 | 1.00 | 0.95 | 0.95 | 1.00 | 0.98 | PES | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 | PET | 1.00 | 0.87 | 0.93 | 0.85 | 1.00 | 0.92 | PLA | 0.95 | 1.00 | 0.98 | 1.00 | 0.76 | 0.86 | PMMA | 1.00 | 0.94 | 0.97 | 0.94 | 1.00 | 0.97 | POM | 1.00 | 0.83 | 0.90 | 0.88 | 0.96 | 0.92 | PP | 0.95 | 1.00 | 0.97 | 0.90 | 1.00 | 0.95 | PPO | 1.00 | 1.00 | 1.00 | 0.84 | 0.89 | 0.86 | PPS | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 | PS | 0.90 | 0.90 | 0.90 | 0.95 | 0.90 | 0.93 | PTFE | 1.00 | 0.95 | 0.97 | 1.00 | 0.95 | 0.97 | PVA | 1.00 | 1.00 | 1.00 | 1.00 | 0.95 | 0.97 | PVC | 0.94 | 1.00 | 0.97 | 0.94 | 1.00 | 0.97 | SBS | 1.00 | 1.00 | 1.00 | 0.95 | 1.00 | 0.98 | TPU | 1.00 | 1.00 | 1.00 | 1.00 | 0.81 | 0.90 |
|
Table 1. Comparison of operation results between XGBoost and SVM algorithms