Research Article
Multitype Damage Detection of Container Using CNN Based on Transfer Learning
Table 3
On-site experiments results: the model outputs, the three prediction types, and probabilities of each image.
| No. | Forecast damage type and probability |
| I | | Norm | | | 100% | | II | Norm | 96.38% | Open | 2.94% | Hole | 0.40% | III | Rusty | 40.51% | Damage | 37.54% | Open | 12.91% | IV | Damage | 81.99% | Bent | 12.51% | Open | 4.50% | V | Damage | 80.14% | Dent | 17.06% | Bent | 1.86% | VI | Damage | 88.47% | Collapse | 6.23% | Dent | 2.58% | VII | Dent | 96.13% | Damage | 2.53% | Hole | 1.25% | VIII | Dent | 68.71% | Damage | 16.52% | Bent | 11.26% | IX | Dent | 88.64% | Collapse | 11.31% | Bent | 0.03% |
|
|