Weld Defect Detection Method Based on Improved YOLOv9

Shengjun Xu; Yiheng Hu; Erhu Liu; Ya Shi; Xiaohan Li; Zongfang Ma

doi:10.3788/LOP250568

Laser & Optoelectronics Progress, Volume. 62, Issue 16, 1622003(2025)

Weld Defect Detection Method Based on Improved YOLOv9

Shengjun Xu^1,2, Yiheng Hu^1,2、*, Erhu Liu^1,2, Ya Shi^1,2, Xiaohan Li^1,2, and Zongfang Ma^1,2

¹College of Information and Control Engineering, Xi’an University of Architecture and Technology, Xi’an 710055, Shaanxi , China

²Key Laboratory of Intelligent Automation Technology for Building Manufacturing, Xi’an 710055, Shaanxi , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(7)

Fig. 1. DA-YOLO network framework diagram

Download full size

Fig. 2. Structure diagram of DSAM

Download full size

Fig. 3. Structure diagram of ADCM

Download full size

Fig. 4. Images of different defect types from the WELD-DETECT data set. (a) Concave; (b) burr; (c) porosity

Download full size

Fig. 5. Detection effect visualization of different networks

Download full size

Table 1. Experimental results of the precision comparison between the proposed network and other networks on the WELD-DETECT defect dataset

View table

Table 1. Experimental results of the precision comparison between the proposed network and other networks on the WELD-DETECT defect dataset

Method	Parameters /10⁶	mAP@0.5 /%	Detection speed /（frame /s）	AP /%
Method	Parameters /10⁶	mAP@0.5 /%	Detection speed /（frame /s）	Burr	Concave	Porosity
Faster R-CNN	110.90	39.8	14.9	42.5	38.3	39.8
MobileNet	4.70	35.7	13.6	57.2	25.6	25.3
EfficientDet	6.60	41.8	13.4	11.5	90.5	23.3
SSD	24.01	50.2	16.5	28.9	73.8	47.9
YOLOv5	7.07	48.8	18.3	60.5	36.2	50.9
YOLOv7	37.21	48.2	16.7	64.7	36.3	45.4
YOLOv8	12.15	45.2	15.4	61.7	33.3	42.4
YOLOv9	60.80	51.2	20.3	65.4	39.7	48.3
Proposed	62.21	56.1	25.8	66.5	45.8	55.9

Table 2. Ablation experiment results

View table

Table 2. Ablation experiment results

Module	mAP@0.5 /%	Parameters /10⁶	Detection speed /（frame /s）
Baseline	51.2	60.80	20.3
Baseline+DSAM	53.7（+2.5）	61.81	22.7
Baseline+ADCM	53.1（+1.9）	61.15	22.4
Baseline+MPDIoU	51.8（+0.6）	60.80	20.9
Baseline+DSAM+ADCM	55.7（+4.5）	62.21	24.6
Baseline+DSAM+ADCM+MPDIoU	56.1（+4.9）	62.21	25.8

Tools

Get Citation

Copy Citation Text

Shengjun Xu, Yiheng Hu, Erhu Liu, Ya Shi, Xiaohan Li, Zongfang Ma. Weld Defect Detection Method Based on Improved YOLOv9[J]. Laser & Optoelectronics Progress, 2025, 62(16): 1622003

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Optical Design and Fabrication

Received: Jan. 23, 2025

Accepted: Mar. 5, 2025

Published Online: Aug. 8, 2025

The Author Email: Yiheng Hu (1214119126@qq.com)

DOI:10.3788/LOP250568

CSTR:32186.14.LOP250568

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology