Research Article
Deep Learning Technology for Weld Defects Classification Based on Transfer Learning and Activation Features
Table 5
Performance metrics of DCNN models with processing time.
| Pretrained model | Accuracy | Error | Sensitivity | Specificity | Precision | False positive rate | Time |
| AlexNet | 100 | 0 | 100 | 100 | 100 | 0 | 47 s | VGG-16 | 95 | 0.05 | 100 | 85 | 92 | 0.145 | 3 min 34 s | VGG-19 | 97.8 | 0.0216 | 100 | 93.75 | 96.8 | 0.0625 | 3 min 55 s | GoogLeNet | 99.3 | 0.0072 | 98.9 | 100 | 100 | 0 | 1 min 38 s | ResNet50 | 100 | 0 | 100 | 100 | 100 | 0 | 3 min 12 s | ResNet101 | 100 | 0 | 100 | 100 | 100 | 0 | 6 min 52 s |
|
|