Research Article
A Novel Multicategory Defect Detection Method Based on the Convolutional Neural Network Method for TFT-LCD Panels
Table 4
Sensitivity and specificity results—original dataset.
| Label in the model | Category | Specificity (%) | Sensitivity (%) | TPA (%) |
| 0 | Blue full-dark defect | 99.37 | 88.04 | 96.60 | 1 | Blue half-dark defect | 99.37 | 89.13 | 2 | Blue half-bright defect | 98.12 | 77.38 | 3 | Blue full-bright defect | 98.38 | 76.74 | 4 | Blue defect-free | 98.12 | 83.13 | 5 | Green full-dark defect | 100 | 94.74 | 6 | Green half-dark defect | 99.82 | 95.38 | 7 | Green half-bright defect | 99.82 | 95.12 | 8 | Green full-bright defect | 99.65 | 96.92 | 9 | Green defect-free | 99.55 | 100 | 10 | Red full-dark defect | 99.38 | 86.57 | 11 | Red half-dark defect | 98.92 | 91.67 | 12 | Red half-bright defect | 99.38 | 87.01 | 13 | Red full-bright defect | 100 | 86.52 | 14 | Red defect-free | 99.91 | 100 |
|
|