Optics and Precision Engineering, Volume. 31, Issue 3, 404(2023)
Defect detection of cylindrical surface of metal pot combining attention mechanism
Figures & Tables(17)
Fig. 7. Samples of various defects on cylindrical surface of metal pot
Fig. 9. Statistics of defect recall rate of each category under different
Fig. 10. Detection results of cylindrical surface defects of highlight reflective metal pot
Table 1. Model super parameter setting
View tableView in ArticleTable 1. Model super parameter setting
Super parameter Value Batch size 4 Cosine scheduler 0.000 1 Weight decay 0.000 01 NMS 0.05 Epoch 300
Table 2. Performance comparison of feature fusion network
View tableView in ArticleTable 2. Performance comparison of feature fusion network
Neck Params
/M
FLOPs
/G
FPS
/（frame∙s-1）
mAP0.5
/%
PANet 9.0 29.977 28.00 86.52 BiFPN 6.4 25.924 32.20 90.06
Table 3. Comparison of detection effect of regression loss functions
View tableView in ArticleTable 3. Comparison of detection effect of regression loss functions
Reg_Loss FPS/（frame∙s-1） mAP0.5/% mAP0.75/% IoU 32.10 85.20 36.20 GIoU 32.00 86.11 38.47 DIoU 32.00 88.94 35.29 CIoU 32.00 89.30 39.63
Table 4. Comparison of detection effect of classification loss functions
View tableView in ArticleTable 4. Comparison of detection effect of classification loss functions
Obj_Loss α γ θ mAP0.5/% mAP0.75/% FL 0.75 1 - 90.02 34.99 FL 0.75 2 - 89.30 33.63 SFL 0.75 1 0.25π 90.14 38.95 SFL 0.75 1 0.5π 91.37 36.56 SFL 0.75 1 0.75π 90.91 35.31 SFL 0.75 1 π 90.58 35.01
Table 5. Performance comparison of feature fusion networks with Di_CBAM introduced at different locations
View tableView in ArticleTable 5. Performance comparison of feature fusion networks with Di_CBAM introduced at different locations
Index Location FNF FPS/
（frame∙s-1）
mAP0.5
/%
mAP0.75
/%
Add1 Add2 Add3 Add4 1 0 0 0 0 1 32.22 89.57 33.28 2 1 1 0 0 1 29.98 89.66 35.68 3 0 0 1 1 1 30.84 90.92 38.21 4 0 1 1 1 1 29.71 89.69 34.57 5 1 1 1 1 1 27.70 89.59 35.98 6 1 1 1 1 0 27.78 88.46 33.52
Table 6. Comparison of impact of improvement strategies on model performance
View tableView in ArticleTable 6. Comparison of impact of improvement strategies on model performance
Index BiFPN Di_CBAM SFL FPS/（frame∙s-1） mAP0.5/% mAP0.75/% 1 0 0 0 28.00 86.52 39.07 2 1 0 0 32.20 90.06 39.53 3 0 1 0 27.60 88.13 39.54 4 0 0 1 28.00 86.89 38.86 5 1 1 0 30.84 90.38 39.65 6 1 0 1 32.00 91.37 36.56 7 0 1 1 27.60 88.26 38.97 8 1 1 1 30.84 90.92 38.21
Table 7. Performance comparison of lightweight improved models
View tableView in ArticleTable 7. Performance comparison of lightweight improved models
Model Params/M FPS/（frame∙s-1） mAP0.5/% mAP0.75/% Light-YOLOv3 8.6 28.70 86.12 35.92 Light-YOLOv5 7.3 29.06 87.35 37.48 YOLOX 9.0 28.00 86.52 39.07 BiYOLOX（Di_CBAM+SFL） 6.4 30.84 90.92 38.21
Tools
Get Citation
Copy Citation Text
Jian QIAO, Nengda CHEN, Yanxiong WU, Yang WU, Jingwei YANG. Defect detection of cylindrical surface of metal pot combining attention mechanism[J]. Optics and Precision Engineering, 2023, 31(3): 404
Paper Information
Category: Information Sciences
Received: Jun. 1, 2022
Accepted: --
Published Online: Mar. 7, 2023
The Author Email: YANG Jingwei (mejwyang@fosu.edu.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20