Welding-Stud Detection Method Based on Depth Perception and Multi-Scale Feature Fusion

Table 1. Results of camera calibration
View table
Table 1. Results of camera calibration
Intrinsic parameters Calibration results
RGB camera /pixel Depth camera /pixel
$f_{x}$ 604.797 385.678
$f_{y}$ 604.589 385.678
$u_{0}$ 317.509 321.942
$v_{0}$ 246.971 237.924

Table 2. Experimental configuration table

View table

Table 2. Experimental configuration table

Parameter	Configuration
Operating system	Windows11
Programming language	Python3.10
Deep learning framework	PyTorch2.1.0
CPU	AMD Ryzen 7 7840H
GPU	NVIDIA GeForce RTX 4060
Graphics card acceleration	CUDA11.8
RAM/Gbit	16

Table 3. Experimental training parameter settings
View table
Table 3. Experimental training parameter settings
Parameter Configuration
Image size /（pixel×pixel） 640 $\times$ 640
Epoch 100
Batch size 4
Learning rate 0.0001

Table 4. Ablation experiment

View table

Table 4. Ablation experiment

Model	Module	Recall /%	mAP@0.5 /%	mAP@0.50∶0.95 /%	Parameter /MBit	GFLOPs	FPS
M1	None	80.0	84.0	54.1	38.6	56.9	63.5
M2	FasterEMA	84.1	86.5	52.5	33.0	51.4	52.4
M3	FasterEMA+TECGA	83.5	86.9	54.2	32.7	51.5	53.2
M4	FasterEMA+TECGA+SN-CCFM	86.6	88.2	56.2	31.3	47.8	45.3

Table 5. Comparative experiments of different algorithms

View table

Table 5. Comparative experiments of different algorithms

Model	mAP@0.5/%	Parameter /MBit	GFLOPs	FPS
RT-DETR	84.0	38.6	56.9	63.5
DETR	76.9	41.6	72.7	28.9
DINO	85.3	47.5	216.1	8.8
Faster R-CNN	79.0	41.3	161.8	18.1
YOLOv5	84.2	21.1	22.6	96.3
YOLOv8	84.1	24.9	27.2	94.8
Proposed model	88.2	31.3	47.8	45.3

Tools

Get Citation

Copy Citation Text

Kaiqi Huang, Chenkang Jin. Welding-Stud Detection Method Based on Depth Perception and Multi-Scale Feature Fusion[J]. Laser & Optoelectronics Progress, 2025, 62(4): 0412006

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites