Weld Image Detection and Recognition Based on Improved YOLOv4

Algorithm	Pore		Slag inclusion		Incomplete penetration		Detection speed /（frame·s^-1）
Algorithm	Precision /%	Recall /%	Precision /%	Recall /%	Precision /%	Recall /%	Detection speed /（frame·s^-1）
YOLOv4-cs	95	74.43	96.52	84.74	84.2	67.59	24.47
YOLOv4	97.13	56.23	95.6	75.47	87.02	53.42	20.43
YOLOv3	96.58	66.27	98.09	80.95	94.78	48.41	25
YOLOv4-Tiny	92.29	60.27	93.18	70.42	85.05	20.4	84.7
SSD	96.18	44.5	97.19	61.89	99.95	0.11	34.1

Table 2. Comparison of F1 values of different models
View table
Table 2. Comparison of F1 values of different models
Model Pore F1 Slag inclusion F1 Incomplete penetration F1
YOLOv4-cs 0.83 0.9 0.75
YOLOv4 0.71 0.84 0.66
YOLOv3 0.79 0.89 0.64
YOLOv4-Tiny 0.73 0.8 0.33
SSD 0.61 0.76 0.002

Table 3. Comparison of AP and mAP values in different models
View table
Table 3. Comparison of AP and mAP values in different models
Model Pore AP /% Slag inclusion AP /% Incomplete penetration AP /% mAP /%
YOLOv4-cs 91 95 80 88.52
YOLOv4 87 92 79 85.85
YOLOv3 88 95 76 86.59
YOLOv4-Tiny 74 83 44 67.02
SSD 82 90 41 71.06

Tools

Get Citation

Copy Citation Text

Song Cheng, Jintao Dai, Honggang Yang, Yunxia Chen. Weld Image Detection and Recognition Based on Improved YOLOv4[J]. Laser & Optoelectronics Progress, 2022, 59(16): 1610002

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites