Surface-Mount Technology Production Line Component Detection System Based on Improved YOLOv5s

Dongdong Wei; Yang Li; Chengzong Yuan

doi:10.3788/LOP250692

Laser & Optoelectronics Progress, Volume. 62, Issue 12, 1212006(2025)

Surface-Mount Technology Production Line Component Detection System Based on Improved YOLOv5s

Dongdong Wei^1、*, Yang Li^1、**, and Chengzong Yuan²

Author Affiliations

¹Shanghai Aerospace Computer Technology Institute, Shanghai 201109, China

²Shanghai Spaceflight Institute of TT&C and Telecommunication, Shanghai 201109, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(13)

Fig. 1. Flowchart of the system design

Download full size

Fig. 2. Line-scan camera and motion control system. (a) Outside view; (b) inside view

Download full size

Fig. 3. Improved YOLOv5s network architecture

Download full size

Fig. 4. Diagram of the backbone network incorporating 4 different attention modules

Download full size

Fig. 5. Diagram of the CA structure module

Download full size

Fig. 6. Examples of typical tray specifications. (a) 10×12; (b) 6×6; (c) 9×17; (d) 5×16

Download full size

Fig. 7. Examples of typical chip specifications. (a) Rectangular; (b) rectangular; (c) square; (d) rectangular

Download full size

Fig. 8. Tray chip detection results. (a) Original image; (b) tray detection and segmentation; (c) chip recognition

Download full size

Fig. 9. Application system user interface

Download full size

Table 1. Analysis of detection results for different specification tray areas

View table

Table 1. Analysis of detection results for different specification tray areas

Type	Size /（mm×mm）	Number of rows	Number of columns	$P$	Type	Size /（mm×mm）	Number of rows	Number of columns	$P$
1	147×125	17	14	0.95	13	135×154	10	12	1.00
2	142×126	12	10	0.96	14	142×128	11	11	0.98
3	135×116	9	8	0.99	15	62×59	20	10	0.97
4	147×128	9	10	0.98	16	63×10	20	2	0.96
5	46×115	4	10	1.00	17	62×22	20	4	1.00
6	26×127	3	14	1.00	18	62×35	20	6	1.00
7	70×130	8	14	0.97	19	77×129	10	2	0.98
8	33×125	3	10	1.00	20	62×22	10	3	1.00
9	63×118	4	8	1.00	21	63×31	10	4	1.00
10	142×126	12	10	0.97	22	63×64	6	6	1.00
11	145×128	9	17	0.96	23	40×62	5	16	1.00
12	145×125	10	10	0.95	24	62×60	20	10	0.98

Table 2. Comparison of experimental results with different attention mechanisms
View table
Table 2. Comparison of experimental results with different attention mechanisms
Model $R$ $P$ mAP $N_{F P S}$ /（frame/s）
YOLOv5s 0.934 0.936 0.956 0.8
YOLOv5s-SE 0.936 0.945 0.971 0.7
YOLOv5s-CBAM 0.945 0.943 0.968 0.45
YOLOv5s-ECA 0.940 0.944 0.965 0.6
YOLOv5s-CA 0.946 0.948 0.972 0.65

Table 3. Ablation experiment comparison of CA and P2 optimization schemes
View table
Table 3. Ablation experiment comparison of CA and P2 optimization schemes
Optimization scheme $R$ $P$ mAP $N_{F P S}$ /（frame/s）
CA P2
/ / 0.934 0.936 0.956 0.8
√ / 0.940 0.942 0.960 0.75
/ √ 0.938 0.939 0.958 0.7
√ √ 0.946 0.948 0.972 0.65

Table 4. Comparison results of YOLOv5s-CA with current mainstream algorithms
View table
Table 4. Comparison results of YOLOv5s-CA with current mainstream algorithms
Model $R$ $P$ mAP $N_{F P S}$ /（frame/s）
SSD 0.84 0.87 0.76 0.35
Faster R-CNN 0.88 0.90 0.83 0.1
YOLOv5l 0.93 0.94 0.92 0.6
YOLOv5m 0.91 0.93 0.90 0.7
YOLOv5s-CA 0.946 0.948 0.972 0.65

Tools

Get Citation

Copy Citation Text

Dongdong Wei, Yang Li, Chengzong Yuan. Surface-Mount Technology Production Line Component Detection System Based on Improved YOLOv5s[J]. Laser & Optoelectronics Progress, 2025, 62(12): 1212006

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites