Defect Detection Method for Vehicle Weld Porosity Based on Improved YOLOv5

Experimental environment	Configuration
Operating system	Linux Ubuntu 22.04
Central processing unit （CPU）	Intel（R） Xeon（R） Platinum 8352V CPU @ 2.10 GHz
Graphics processing unit （GPU）	NVIDIA GeForce RTX 4090 （24 GB）
Deep learning framework	PyTorch 2.1.1
Computing platform	CUDA 11.8
Interpreter	Python 3.10.13

Table 2. Parameters for model training
View table
Table 2. Parameters for model training
Parameter Value
Epoch 350
Batch size 64
Minimum learning rate 0.00001
Maximum learning rate 0.01
Resolution /（pixel×pixel×pixel） 640×640×3
Patience 50

Table 3. Results of comparison experiments by different detection methods
View table
Table 3. Results of comparison experiments by different detection methods
Method mAP0.5 /% FPS /（frame·s^-1） Parameters
SSD 71.9 48.82 23745908
YOLOv5s 59.7 100.48 7053910
YOLOv5m 62.3 77.52 20871318
YOLOv8s 49.0 108.70 3011043
Proposed method 62.9 89.72 7291390

Table 4. Results of ablation experiments on WELD-DET dataset
View table
Table 4. Results of ablation experiments on WELD-DET dataset
Method mAP0.5 /% FPS /（frames·s^-1）
Yolov5s 59.7 100.48
Yolov5s+CBAM 61.3 95.82
Yolov5s+BiFPN 61.4 94.05
Yolov5s+ WIoU 60.6 98.04
Yolov5s+BiFPN+CBAM 61.8 91.95
Yolov5s+ BiFPN+CBAM+WIoU 62.9 89.72

Tools

Get Citation

Copy Citation Text

Xiaolong Zhou, Changjie Liu. Defect Detection Method for Vehicle Weld Porosity Based on Improved YOLOv5[J]. Laser & Optoelectronics Progress, 2025, 62(4): 0412005

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites