Fig 1..
The metallized ceramic ring and the pinhole defects on the surface of its metal layer.
Table 1..
Model and parameters of each device in the acquisition device.
Fig 2.
Schematic diagram of the pinhole defect detection module of the metallized ceramic ring.
Fig 3.
Sample image preprocessing and image segmentation process.
Table 2.
Numbers of pinhole defects before and after data augmentation.
Fig 4.
Basic framework of data augmentation.
Fig 5.
Improved DETR Network Model.
Fig 6.
Principles of SimAM attention mechanism.
Fig 7.
MBconv structure with SimAM.
Fig 8.
Pinhole defect detection procedure based on deep learning.
Fig 9.
The morphology-based pinhole defect detection module flow.
Fig 10.
Verification flow of detection results of the two pinhole defect detection modules.
Fig 11.
Fusion process of detection results of the two pinhole defect detection modules.
Fig 12.
Implementation process chart of proposed method.
Table 3.
Training parameters of the improved DETR network model.
Table 4.
mAP of the model before and after data augmentation.
Table 5.
Settings of different models and comparison of detection results.
Table 6.
Pinhole Detection Results of Different Methods.
Table 7.
Comparison of performances of different detection methods.
Table 8.
Single-frame image inference time of different detection methods.