Laser & Optoelectronics Progress, Volume. 60, Issue 20, 2015007(2023)
Defect Detection of Metallized-Ceramic Rings Based on Fusion of Object Detection and Image Classification Networks
Figures & Tables(11)
Fig. 2. Defect category. (a) Black spot; (b) lack; (c) pulp point; (d) pinhole; (e) fall off
Fig. 4. Defect category distribution before and after data augmentation. (a) Defect category distribution before data augmentation; (b) defect category distribution after data augmentation
Fig. 8. Effect comparison of different defect detection models. (a) YOLOv3; (b) SSD; (c) Faster-RCNN(VGG-16); (d) Faster-RCNN(ResNet-50+FPN); (e) proposed method
Table 1. Comparison of the precision and computing resource requirement between different methods
View tableTable 1. Comparison of the precision and computing resource requirement between different methods
Parameter YOLOv3 SSD Faster-RCNN（VGG-16） Faster-RCNN（ResNet-50+FPN） Proposed method AP/% Pulp point 66.7 78.3 81.2 82.3 96.1 Pinhole 0 6.8 22.9 75.3 80.5 Black spot 50.6 90.6 91.2 94.8 100 Lack 57.1 47.3 56.4 61.0 68.9 Fall off 36.7 34.9 39.4 65.1 76.7 mAP /% 42.2 51.6 58.2 75.7 84.4 Parameters /106 62.6 13.7 43.9 41.3 64.8 Volume /MB 323 105 334 315 405 Inference time /ms 16.1 18.9 69.9 74.8 146.4
Table 2. Comparison of the precision of different networks in network fusion architecture
View tableTable 2. Comparison of the precision of different networks in network fusion architecture
Parameter Faster-RCNN（ResNet-50+FPN）+ VGG Faster-RCNN（ResNet-50+FPN）+ AlexNet Proposed method AP/% Pulp point 86.6 85.1 96.1 Pinhole 77.2 76.6 80.5 Black spot 100 98.9 100 Lack 63.6 65.1 68.9 Fall off 68.4 69.6 76.7 mAP/% 79.2 79.1 84.4
Table 3. Comparison of the precision and recall of two methods
View tableTable 3. Comparison of the precision and recall of two methods
Defect category Faster-RCNN（ResNet-50+FPN） Proposed method Precision Recall Precision Recall Pulp point 94.2 67.3 98.4 82.4 Pinhole 85.2 68.2 94.6 70.1 Black spot 100 62.3 100 62.7 Lack 82.3 73.1 91.6 90.0 Fall off 33.8 84.5 76.3 82.3
Tools
Get Citation
Copy Citation Text
Yingjie Man, Xian Wang, Dongyue Sun, Ningdao Deng, Shixu Wu. Defect Detection of Metallized-Ceramic Rings Based on Fusion of Object Detection and Image Classification Networks[J]. Laser & Optoelectronics Progress, 2023, 60(20): 2015007
Paper Information
Category: Machine Vision
Received: Nov. 7, 2022
Accepted: Dec. 23, 2022
Published Online: Sep. 28, 2023
The Author Email: Xian Wang (15111388435@163.com)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20