Research Article
Deep Learning Technology for Weld Defects Classification Based on Transfer Learning and Activation Features
Table 8
Comparison of the performance metrics of pretrained DCNN models.
| ā | Accuracy (%) | TP (%) | TN (%) | FN (%) | FP (%) |
| Our method | 85.2 | 39.8 | 54.4 | 2.8 | 12.0 | VGG-16 | 84.3 | 37 | 47.2 | 5.6 | 10.2 | VGG-19 | 80.6 | 25.9 | 54.6 | 16.7 | 2.2 | GoogLeNet | 74.1 | 21.3 | 52.8 | 21.3 | 4.6 | ResNet50 | 68.5 | 29.6 | 38.9 | 13.0 | 18.5 | ResNet101 | 74 | 25.0 | 49.1 | 17.6 | 8.3 | DCFA | 71.3 | 22.2 | 49.1 | 20.4 | 8.3 |
|
|