TABLE 3.
The performance of 10 deep learning-based models in the test set.
| Accuracy of pattern classifications |
|||||
| Models | Normal (%) | Obstruction (%) | Restriction (%) | Mixed (%) | Average accuracy (%) |
| ResNet18 | 98.3 | 88.2 | 94.8 | 97.3 | 94.7 |
| ResNet34 | 97.0 | 87.1 | 94.8 | 91.7 | 92.7 |
| ResNet18_vd | 98.03 | 88.9 | 94.1 | 95.4 | 94.2 |
| ResNet34_vd | 98.5 | 90.3 | 95.2 | 95.9 | 95.0 |
| ResNet50_vd | 98.8 | 87.3 | 94.8 | 95.4 | 94.1 |
| ResNet50_vc | 98.1 | 86.2 | 93.0 | 96.3 | 93.4 |
| SE_ResNet18_vd | 97.7 | 86.9 | 96.3 | 93.6 | 93.6 |
| VGG11 | 98.2 | 91.0 | 96.7 | 94.5 | 95.1 |
| VGG13 | 97.7 | 91.0 | 96.7 | 97.3 | 95.6 |
| VGG16 | 98.4 | 92.3 | 95.9 | 95.0 | 95.4 |