Research Article
Identification and Classification of Atmospheric Particles Based on SEM Images Using Convolutional Neural Network with Attention Mechanism
Table 3
The four evaluation indexes of Attention-CNN, CNN, and SVM.
| Classification type | Models | Precision (%) | Recall (%) | Specificity (%) | F1-score (%) |
| Fibrous | Attention-CNN | 98.33 | 96.71 | 99.32 | 97.51 | CNN | 97.66 | 94.81 | 99.05 | 96.21 | SVM | 92.31 | 92.93 | 96.91 | 92.62 | Flocculent | Attention-CNN | 96.67 | 98.31 | 98.66 | 97.48 | CNN | 88.00 | 97.78 | 95.33 | 92.63 | SVM | 86.67 | 86.09 | 94.59 | 86.38 | Spherical | Attention-CNN | 94.17 | 89.74 | 98.39 | 91.90 | CNN | 87.00 | 92.82 | 96.51 | 89.81 | SVM | 86.10 | 91.87 | 96.27 | 88.89 | Mineral | Attention-CNN | 84.02 | 88.46 | 95.80 | 86.18 | CNN | 89.50 | 79.35 | 97.10 | 84.12 | SVM | 72.15 | 67.81 | 92.45 | 69.91 |
|
|
Bold values indicate the best results of the four indexes in the three models.
|