An Improved YOLOv5 Algorithm for Steel Surface Defect Detection

Fig. 7. Detection results of the proposed model on NEU-DET dataset. (a) Crazing; (b) inclusion; (c) patches; (d) pitted surface; (e) rolled-in scale; (f) scratches

Download full size

Table 1. Experimental results of different models on NEU-DET dataset

View table

Table 1. Experimental results of different models on NEU-DET dataset

Method	AP（Cr）/%	AP（In）/%	AP（Pa）/%	AP（Ps）/%	AP（Rs）/%	AP（Sc）/%	mAP /%	FPS /（frame/s）
Faster R-CNN^［3］	35.96	81.80	88.34	81.82	59.31	86.53	72.29	7.56
SSD^［4］	38.50	80.78	93.94	82.62	67.96	72.13	72.66	43.24
RetinaNet^［5］	36.38	81.62	91.56	81.74	60.81	88.10	73.37	30.26
YOLOv3^［6］	34.26	83.88	91.52	81.74	60.91	88.32	73.44	43.67
YOLOv4^［7］	31.42	84.61	94.72	81.60	61.36	92.86	74.43	31.15
YOLOv5-s	39.13	79.21	96.46	87.12	60.19	84.19	74.38	62.75
YOLOv5-m	38.96	84.50	94.81	87.59	61.31	89.03	76.03	48.07
YOLOX^［18］	34.19	84.67	97.59	88.72	67.36	90.41	77.28	47.25
YOLOv7^［27］	38.97	81.60	93.74	81.61	62.73	83.80	73.74	44.72
YOLOv8	41.21	84.66	93.97	88.43	63.17	96.66	78.01	61.33
MSC-DNet^［8］	42.40	84.50	94.30	91.50	71.60	92.00	79.38	14.10
DEA_RetinaNet^［9］	60.93	82.49	94.27	95.79	67.16	74.05	79.11	12.20
ES-Net^［11］	56.00	87.60	88.30	87.40	60.40	94.90	79.10	－
This Research	43.31	87.24	96.53	88.33	72.80	94.84	80.51	31.96

Table 2. Results of ablation study 1

View table

Table 2. Results of ablation study 1

Method	mAP /%	Improvement /百分点	FPS /（frame/s）
YOLOv5-m	76.03	－	48.07
YOLOv5-m + RFB	78.07	2.04	41.08
YOLOv5-m+AFAM	77.05	1.02	43.27
YOLOv5-m+decoupled head without ECA	77.46	1.43	39.81
YOLOv5-m+decoupled head with ECA	77.94	1.91	39.35

Table 3. Results of ablation study 2
View table
Table 3. Results of ablation study 2
RFB AFAM Decoupled head with ECA mAP /% Improvement /百分点 FPS /（frame/s）
－－－ 76.03 － 48.07
√ √ － 78.81 2.78 35.72
√ － √ 79.62 3.59 33.19
－ √ √ 78.73 2.70 36.28
√ √ √ 80.51 4.48 31.96

Tools

Get Citation

Copy Citation Text

Shaoxiong Li, Zaifeng Shi, Fanning Kong, Ruoqi Wang, Tao Luo. An Improved YOLOv5 Algorithm for Steel Surface Defect Detection[J]. Laser & Optoelectronics Progress, 2023, 60(24): 2412003

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites