Fig 1.
(a) missing hole; (b) mouse bite; (c) open circuit; (d) short; (e) spur; (f) spurious copper.
Fig 2.
PCB dataset augmentation methods.
(a) original figure; (b) noise injection; (c) brightness variation; (d) rotating; (e) cutout, (f) cropping.
Fig 3.
Structure of YOLOv10n network.
Fig 4.
Structure of EDF-YOLOv10n network.
Fig 5.
Structure of ECA module.
Fig 6.
Structure of C2f and DSCC2f module.
(a) structure of C2f module; (b) structure of DSCC2f module.
Fig 7.
Structure of DSConv module.
Table 1.
Comparison of detection accuracies between EDF-YOLOv10 and YOLOv10.
Fig 8.
Comparison of EDF-YOLOv10 and YOLOv10 indices.
(a) precision; (b) recall; (c) mAP@0.50; (d) mAP@0.50:0.95.
Fig 9.
Comparison of confusion matrices between YOLOv10 and EDF-YOLOv10.
(a) YOLOv10; (b) EDF-YOLOv10.
Fig 10.
(a) missing hole; (b) mouse bite; (c) spur; (d) spurious copper.
Fig 11.
Detection results of EDF-YOLOv10.
(a) missing hole; (b) mouse bite; (c) spur; (d) spurious copper.
Table 2.
Comparison of detection performance between YOLOv10 and EDF-YOLOv10 from five experiments.
Table 3.
Results of the ablation experiment.
Table 4.
Performance comparison of different loss functions for YOLOv10 model with ECA and DSConv enhancements.
Table 5.
Comparative experimental results for different models on the PKU-Market-PCB dataset.
Table 6.
Comparative experimental results for different models on the DeepPCB dataset.
Fig 12.
PCB simulation experimental results.
(a) hardware system; (b) detection results on the visualization interface.