Table 8 Defect detection performance of different methods.
From: Enhanced PCB defect detection via HSA-RTDETR on RT-DETR
Model | Backbone | P/% | R/% | mAP50/% | mAP95/% | Params/m |
|---|---|---|---|---|---|---|
Faster R-CNN | ResNet50 | 89.7 | 86.6 | 90.6 | 40.3 | 42 |
CornerNet | Hourglass | 85.1 | 83.3 | 87.5 | 37.5 | 38.4 |
YOLOv5s | – | 90.6 | 90.7 | 93.2 | 45 | 7.1 |
YOLOv7 | – | 92.4 | 96.1 | 96.4 | 47.6 | 34.8 |
YOLOv8n | – | 96.9 | 90.9 | 96.3 | 49.7 | 25.9 |
DETR | ResNet50 | 87.6 | 85.4 | 89.3 | 40.2 | 41.3 |
ResNet101 | 89.1 | 87.7 | 90.7 | 41.1 | 74.3 | |
ResNet18 | 85.5 | 83.1 | 86.4 | 38.7 | 23.6 | |
Deformable DETR | ResNet50 | 86.5 | 88.7 | 91.1 | 40.7 | 40 |
ResNet101 | 89.2 | 90.2 | 91.6 | 42.2 | 75.4 | |
ResNet18 | 85.3 | 86.5 | 88.1 | 39.6 | 20.3 | |
Focus-DETR | ResNet50 | 92.1 | 87.6 | 90.7 | 42.8 | 48 |
ResNet101 | 93.3 | 89.3 | 92.2 | 43.3 | 81.2 | |
ResNet18 | 91.2 | 86.1 | 87.6 | 40.1 | 26.7 | |
RT-DETR | ResNet50 | 91.9 | 88.7 | 93.8 | 45 | 42.8 |
ResNet101 | 93.8 | 91.7 | 94.7 | 42.7 | 76.7 | |
ResNet18 | 91.3 | 86 | 91.5 | 40.8 | 20.1 | |
HSA-RTDETR | R18-Faster-EMA | 97.1 | 93.9 | 96.9 | 45.9 | 17.6 |
