Researching | Defect detection of curved optical lenses based on improved YOLOv5s

Journal of Applied Optics, Volume. 45, Issue 4, 781(2024)

Defect detection of curved optical lenses based on improved YOLOv5s

Xiaolei LIU and Fenghui LIU^*

Author Affiliations

School of Physics and Electronic Information, Henan Polytechnic University, Jiaozuo 454003, China

show less

Abstract Get PDF(in Chinese)

Figures&Tables (15)

References (19)

Paper Information

Figures & Tables(15)

Fig. 1. Images of lens defects

Download full size

View in Article

Fig. 2. Defect collection device

Download full size

View in Article

Fig. 3. Defect images generated by DCGAN

Download full size

View in Article

Fig. 4. Oversampling operation

Download full size

View in Article

Fig. 5. Structure diagram of YOLOv5s network

Download full size

View in Article

Fig. 6. SE module

Download full size

View in Article

Fig. 7. C3TR module

Download full size

View in Article

Fig. 8. Structure diagram of strengthen feature fusion

Download full size

View in Article

Fig. 9. Structure diagram of improved YOLOv5s model

Download full size

View in Article

Fig. 10. Comparison of mean average precision

Download full size

View in Article

Fig. 11. Comparison of defect detection results

Download full size

View in Article

Table 1. Experimental environment
View table
View in Article
Table 1. Experimental environment
操作系统显卡 CUDA CUDNN 网络框架编程语言
Ubuntu RTX3090 11.0 8.0.4 Pytorch1.7.0 Python3.7

Table 2. YOLOv5s ablation experiment results

View in Article

Table 2. YOLOv5s ablation experiment results

模型	$P / %$	$R / %$	$m A P / %$	$A P / %$
模型	$P / %$	$R / %$	$m A P / %$	x	s	h	y
YOLOv5s	91.2	88.2	90.7	91.1	88.0	92.8	90.9
YOLOv5s+SE	91.6	89.0	91.9	92.7	90.0	93.2	91.7
YOLOv5s+SE+TR	92.0	90.4	92.4	93.4	90.8	93.6	91.8
YOLOv5s+SE+C2	91.7	89.7	92.9	94.2	90.5	94.1	92.8
YOLOv5s+SE+TR+C2	92.2	91.6	93.9	94.4	92.1	95.0	94.1

Table 3. Comparison of improved network experiment results
View table
View in Article
Table 3. Comparison of improved network experiment results
模型 YOLOv5s YOLOv8s
$P / %$ $R / %$ $m A P / %$ $P / %$ $R / %$ $m A P / %$
原始 91.2 88.2 90.7 91.1 85.4 90.0
SE+TR+C2 92.2 91.6 93.9 92.9 86.5 91.5

Table 4. Comparative experimental results of different networks

View in Article

Table 4. Comparative experimental results of different networks

模型	$t / ms$	$m A P / %$	$A P / %$
模型	$t / ms$	$m A P / %$	x	s	h	y
Mask R-CNN	41	77.5	76.7	81.2	67.5	84.7
Faster R-CNN	37	74.9	60.4	86.4	78.6	74.2
SSD	21	75.1	88.0	64.1	87.2	61.1
改进YOLOv5s	14	93.9	94.4	92.1	95.0	94.1
YOLOv7s	19	86.1	87.1	85.9	91.3	80.1
YOLOx_s	17	89.5	89.3	88.2	90.5	89.8

Tools

Get Citation

Copy Citation Text

Xiaolei LIU, Fenghui LIU. Defect detection of curved optical lenses based on improved YOLOv5s[J]. Journal of Applied Optics, 2024, 45(4): 781

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Research Articles

Received: Aug. 24, 2023

Accepted: --

Published Online: Oct. 21, 2024

The Author Email: LIU Fenghui (刘丰慧)

DOI:10.5768/JAO202445.0403003

Topics

laser devices and laser physics

Lasers and Laser Optics

laser manufacturing

Instrumentation, Measurement and Metrology