Defect Detection of Tubular Containers Based on an Unsupervised Domain Adaptive Algorithm

Table 1. Comparison of different surface defect after data domain offset
View table
Table 1. Comparison of different surface defect after data domain offset
Defect type Normal Light change Focus change
Scratch
Bubble
Pit
Sticker scratch

Table 2. Number of samples for each defect in the dataset

View table

Table 2. Number of samples for each defect in the dataset

Surface defect	Label	Original		Expanded		Train		Val	Test
Surface defect	Label	Source	Target	Source	Target	Source	Target	Target	Target
Scratch	SR	237	156	300	500	300	300	100	100
Bubble	BU	186	132	300	500	300	300	100	100
Pit	PI	218	182	300	500	300	300	100	100
Sticker Scratch	SS	234	153	300	500	300	300	100	100

Table 3. Comparison experiment
View table
Table 3. Comparison experiment
mAP@0.5 /% Recall /% Precision /%
RCNN 75.6 71.2 77.1
Faster RCNN 81.2 77.8 82.4
YOLOv5 80.6 78.7 81.3
YOLOv8 83.1 79.4 84.2
Proposed 93.4 90.6 93.7
Supervision 95.6 97.3 94.6

Table 4. Ablation experiment

View table

Table 4. Ablation experiment

	IMADA	INSDA	CR	Improved	mAP@0.5 /%	Recall /%	Precision /%
YOLOv8					83.1	79.4	84.2
Proposed	√				86.3	84.6	87.2
	√			√	87.6	85.8	89.2
		√			84.1	82.4	86.7
		√		√	84.6	82.1	88.4
	√	√			88.7	85.6	89.3
	√	√		√	90.2	87.9	91.4
	√	√	√		91.3	89.4	92.1
	√	√	√	√	93.4	90.6	93.7
Supervision					95.6	97.3	94.6

Tools

Get Citation

Copy Citation Text

Guangzhi Zhang, Huimin Li, Xuning Song. Defect Detection of Tubular Containers Based on an Unsupervised Domain Adaptive Algorithm[J]. Laser & Optoelectronics Progress, 2025, 62(2): 0212007

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites