Chinese Journal of Lasers, Volume. 51, Issue 16, 1602103(2024)
Sim-YOLOv8 Object Detection Model for DR Image Defects in Aluminum Alloy Welds
Figures & Tables(11)
Fig. 2. Structures of each module in YOLOv8. (a) CBS module; (b) C2f module; (c) BottleNeck module; (d) SPPF module
Fig. 4. Structure diagrams of improved S-C2f module. (a) S-C2f module; (b) BottleNeck in S-C2f module
Fig. 7. Comparison of model improvement effect. (a) Detection effect of the original model; (b) detection effect of improved model
Table 1. Comparison of precision and performance indexes for different models
View tableTable 1. Comparison of precision and performance indexes for different models
Model AP /% mAP@0.5 /% FPS #Params /106 FLOPs /109 Pore Slag inclusion Incomplete penetration Sim-YOLOv8 93.6 94.4 97.3 95.1 217 3.0 8.1 YOLOv8 91.1 92.5 95.6 93.1 222 3.0 8.1 YOLOv7 Tiny 81.7 90.5 94.1 88.7 222 6.2 13.8 YOLOv5s 88.2 91.1 77.1 85.5 156 7.2 16.5 YOLOv4 Tiny 63.9 77.6 45.6 62.3 102 9.0 26.8 SSD 83 91 49 74.11 208 7.9 17.4 DETR 70 85 26 60.33 159 6.9 6.5
Table 2. Comparison of verification and test sets indicators for YOLOv8
View tableTable 2. Comparison of verification and test sets indicators for YOLOv8
Set AP /% mAP@0.5 /% Pore Slag inclusion Incomplete penetration Test 91.1 92.5 95.6 93.1 Verification 90.1 93.3 95.7 93.0
Table 3. Comparison of verification and test sets indicators for Sim-YOLOv8
View tableTable 3. Comparison of verification and test sets indicators for Sim-YOLOv8
Set AP /% mAP@0.5 /% Pore Slag inclusion Incomplete penetration Test 93.6 94.4 97.3 95.1 Verification 92.5 94.8 96.7 94.7
Table 4. Comparison of different models in terms of parameter number, weight file size and inference speed
View tableTable 4. Comparison of different models in terms of parameter number, weight file size and inference speed
Group SimAM WIoU Focus mAP@0.5 /% FPS #Params FLOPs /109 0 93.1 222 3006233 8.1 1 √ 94.1 （↑ 1 percentage point） 227 3006233 8.1 2 √ √ 94.4 （↑ 1.3 percentage points） 204 3006233 8.1 3 √ √ √ 95.1 （↑ 2 percentage points） 217 3012729 8.1
Tools
Get Citation
Copy Citation Text
Lei Wu, Yukun Chu, Honggang Yang, Yunxia Chen. Sim-YOLOv8 Object Detection Model for DR Image Defects in Aluminum Alloy Welds[J]. Chinese Journal of Lasers, 2024, 51(16): 1602103
Paper Information
Category: Laser Forming Manufacturing
Received: Dec. 7, 2023
Accepted: Feb. 5, 2024
Published Online: Jul. 29, 2024
The Author Email: Yunxia Chen (cyx1978@yeah.net)
CSTR:32183.14.CJL231485
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20