Optics and Precision Engineering, Volume. 32, Issue 10, 1622(2024)
Computer motherboard assembly defect detection using parallel feature extraction and progressive feature fusion
Figures & Tables(15)
Fig. 1. Network architecture for defect detection in computer motherboard assembly
Fig. 3. Network Structure of the VT-FasterBlock Feature Extraction Module
Fig. 5. Multiscale intersection attention asymptotic feature fusion networks
Fig. 8. Examples of types of surface assembly quality defects on computer motherboards
Fig. 10. Computer motherboard assembly defect detection visualization test results
Fig. 11. Heat map of output scores for different structures introduced by the network
Table 1. Comparison of evaluation metrics for different defect detection algorithms
View tableView in ArticleTable 1. Comparison of evaluation metrics for different defect detection algorithms
Method Backbone mAP（@IoU=0.5）/% FPS/s Faster RCNN VGG19 66.4 12 YOLOv3 DarkNet-53 80.8 52 YOLOv4 CSPDarkNet53 83.2 34 YOLOx CSPDarkNet53 89.1 51 YOLOv5 CSPDarkNet53 90.0 56 YOLOv7 ELAN-Net 93.3 19 MobileNetV3-YOLOv4［ 23 ］MobileNetV3 87.1 63 Swin-Transformer-YOLOx［ 24 ］Swin-Transformer 93.1 14 Ours VTNet 94.6 25
Table 2. Results of ablation experiments with the proposed method
View tableView in ArticleTable 2. Results of ablation experiments with the proposed method
Methods mAP（@IoU=0.5）/% AP@（IoU=0.5）/% 1 2 3 4 5 6 7 8 9 10 11 YOLOv5 90.01 94.00 99.66 99.48 98.60 98.66 89.85 40.74 93.40 93.42 83.51 98.80 YOLOv5+改进1 91.74 95.32 98.29 99.28 98.49 97.10 86.92 62.69 94.22 94.89 82.76 98.93 YOLOv5+改进2 91.95 95.40 99.41 98.53 97.63 94.61 79.90 84.79 92.39 94.25 75.26 98.96 YOLOv5+改进1+改进2 94.63 96.41 98.48 98.83 98.19 98.46 88.58 84.38 99.71 96.22 82.65 99.05
Table 3. Experimental results of different feature fusion networks
View tableView in ArticleTable 3. Experimental results of different feature fusion networks
Methods mAP（@IoU=0.5）/% FPS AP@（IoU=0.5）/% 1 2 3 4 5 6 7 8 9 10 11 YOLOv5+PANet 90.01 56 94.00 99.66 99.48 98.60 98.66 89.85 40.74 93.40 93.42 83.51 98.80 YOLOv5+AFPN［ 20 ］91.08 43 92.96 95.96 98.90 98.42 96.36 88.37 68.37 91.54 92.83 79.53 98.94 YOLOv5+Ours 91.95 25 95.40 99.41 98.53 97.63 94.61 79.90 84.79 92.39 94.25 75.26 98.96
Table 4. Experimental results of different feature fusion networks
View tableView in ArticleTable 4. Experimental results of different feature fusion networks
Methods mAP（@IoU=0.5）/% FPS AP@（IoU=0.5）/% 1 2 3 4 5 6 7 8 9 10 11 YOLOv5-ResNet 88.77 17 95.01 98.49 99.27 81.02 95.52 92.96 46.68 93.51 96.95 78.98 98.19 YOLOv5-CSPDarkNet 90.01 56 94.00 99.66 99.48 98.60 98.66 89.85 40.74 93.40 93.42 83.51 98.80 YOLOv5-MobileViT 89.61 41 93.17 98.66 98.99 91.40 95.23 93.02 54.21 93.70 97.25 74.78 95.34 YOLOv5-Swin-Transformer 90.39 16 93.80 96.29 97.06 91.05 95.29 91.41 58.29 93.71 94.26 86.56 96.51 YOLOv5+Ours 91.74 33 95.32 98.29 99.28 98.49 97.10 86.92 62.69 94.22 94.89 82.76 98.93
Tools
Get Citation
Copy Citation Text
Junying CHEN, Zhaoyang LI, Hantao HUANG, Xuze DONG. Computer motherboard assembly defect detection using parallel feature extraction and progressive feature fusion[J]. Optics and Precision Engineering, 2024, 32(10): 1622
Paper Information
Category:
Received: Dec. 8, 2023
Accepted: --
Published Online: Jul. 8, 2024
The Author Email: LI Zhaoyang (nicholas@xauat.edu.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20