Feature Enhancement and Metric Optimization for Defect Detection on Steel Surface

Table 1. Detection results with STCSP module added at different feature extraction layers
View table
Table 1. Detection results with STCSP module added at different feature extraction layers
Number of layer mAP@0.5 AP@0.5
Cr In Pa Ps Rs Sc
— 74.3 35.1 85.0 96.3 80.3 57.2 92.1
24 72.7 34.1 79.3 95.1 80.5 59.2 88.2
37 74.5 36.9 82.7 97.0 83.3 59.1 88.0
50 75.6 37.9 84.1 98.2 81.5 62.0 89.8

Table 2. Detection results with ELAN-DS module added at different positions
View table
Table 2. Detection results with ELAN-DS module added at different positions
Position mAP@0.5 AP@0.5
Cr In Pa Ps Rs Sc
— 74.3 35.1 85.0 96.3 80.3 57.2 92.1
Backbone 76.9 40.9 83.7 97.6 85.1 62.5 91.7
Head 74.1 37.9 83.2 96.4 82.4 58.7 86.0
Backbone & Head 75.5 38.3 83.8 96.0 82.8 61.8 90.1

Table 3. Detection results with different ratios of weighted NWD and CIoU loss function

View table

Table 3. Detection results with different ratios of weighted NWD and CIoU loss function

NWD∶CIoU	mAP@0.5	AP@0.5
NWD∶CIoU	mAP@0.5	Cr	In	Pa	Ps	Rs	Sc
0∶1.0	74.3	35.1	85.0	96.3	80.3	57.2	92.1
0.2∶0.8	76.6	37.3	86.9	97.6	84.6	62.2	91.2
0.8∶0.2	77.8	43.2	85.6	98.1	82.1	64	94.1
1.0∶0	78.7	46.6	84.4	97.8	85.6	64.1	93.5
0.5∶0.5	81.7	50.4	88.8	99.2	86.0	70.6	95.0

Table 4. Results of ablation experiments
View table
Table 4. Results of ablation experiments
Number STCSP ELAN-DS NWD∶CIoU=0.5∶0.5 FPS /（frame·s^-1） mAP@0.5 /%
N1 — — — 82 74.3
N2 + — — 75 75.6
N3 — + — 66 76.9
N4 — — + 84 81.7
N5 + + — 79 74.4
N6 + — + 76 79.3
N7 — + + 77 81.9
N8 + + + 71 83.1

Table 5. Comparison of experimental results for different object detection algorithms

View table

Table 5. Comparison of experimental results for different object detection algorithms

Algorithm	Parameters /M	FPS /（frame·s^-1）	mAP@0.5 /%
Cascade R-CNN	107.0	37	59.6
RetinaNet	28.5	42	61.7
YOLOv5（m）	21.2	75	70.7
Improved YOLOv5	5.88	49	78.5
YOLOX（m）	25.3	88	70.1
CG-Net	32.7	105	75.9
CenterNet	2.3	41	73.4
YOLOv7	36.5	82	74.3
YOLOv8	30.1	118	75.8
YOLOv9	60.5	40	74.4
Proposed	38.8	71	83.1

Tools

Get Citation

Copy Citation Text

Junying Chen, Hantao Huang, Zhaoyang Li. Feature Enhancement and Metric Optimization for Defect Detection on Steel Surface[J]. Laser & Optoelectronics Progress, 2024, 61(24): 2412002

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Instrumentation, Measurement and Metrology

Received: Mar. 12, 2024

Accepted: Apr. 25, 2024

Published Online: Dec. 10, 2024

The Author Email: Junying Chen (chenjy@xauat.edu.cn)

DOI:10.3788/LOP240876

CSTR:32186.14.LOP240876

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Detection results with STCSP module added at different feature extraction layers

Table 1. Detection results with STCSP module added at different feature extraction layers

Table 2. Detection results with ELAN-DS module added at different positions

Table 2. Detection results with ELAN-DS module added at different positions

Table 3. Detection results with different ratios of weighted NWD and CIoU loss function

Table 3. Detection results with different ratios of weighted NWD and CIoU loss function

Table 4. Results of ablation experiments

Table 4. Results of ablation experiments

Table 5. Comparison of experimental results for different object detection algorithms

Table 5. Comparison of experimental results for different object detection algorithms