Metal-YOLO Detection Algorithm for Defects in Coaxial Packaged Metal Base

Fig. 4. Comparison of detection effects between Metal-YOLO and YOLO v5s. (a) True defect; (b) detection effects of YOLO v5s; (c) detection effects of Metal-YOLO

Download full size

Table 1. Ablation experimental results of AAM
View table
Table 1. Ablation experimental results of AAM
Method $m_{m A P_{50}}$ /% FLOPs /G Parameter /M
YOLO v5s 74.2 15.8 14.5
YOLO v5s +AAM（7×7） 75.6 15.8 14.6
YOLO v5s +AAM 76.9 15.8 14.5

Table 2. Comparison of results for different attention mechanisms

View table

Table 2. Comparison of results for different attention mechanisms

Method	$m_{m A P_{50}}$ /%	FLOPs /G	Parameter /M
YOLO v5s+AAM	76.9	15.8	14.5
YOLO v5s+SE	73.0	15.8	14.5
YOLO v5s+ECA	74.9	15.8	14.5
YOLO v5s+CBAM	75.3	15.8	14.5
YOLO v5s+CA	75.5	15.9	14.5
YOLO v5s+GAM	75.6	17.2	17.9
YOLO v5s+EMA	73.9	15.8	14.5
YOLO v5s+SimAM	74.8	15.8	14.5
YOLO v5s+TripletA	75.1	15.9	14.5

Table 3. Detection results of different models after adding AAM
View table
Table 3. Detection results of different models after adding AAM
Method $m_{m A P_{50}}$ /%
YOLO v3tiny 72.2
YOLO v3tiny+AAM 73.3
YOLO v5n 70.8
YOLO v5n+AAM 71.1
YOLO v5m 74.2
YOLO v5m+AAM 75.4
YOLO v5l 73.8
YOLO v5l+AAM 75.1
YOLO v7tiny 53.9
YOLO v7tiny+AAM 56.2

Table 4. Ablation experiment results of Metal-YOLO
View table
Table 4. Ablation experiment results of Metal-YOLO
YOLOv5s CFEC AAM EIoU P /% R /% F_F1 /% $m_{m A P_{50}}$ /% Frame rate /（frame·s^-1）
√ 83.3 69.1 75.5 74.2 37
√ √ 84.3 72.1 77.7 75.9 39
√ √ 84.3 73.1 78.3 76.9 37
√ √ 82.1 72.9 77.2 76.4 40
√ √ √ √ 83.9 74.1 78.7 78.3 34

Table 5. Comparison of experimental results for different detection models on the ILS-MB dataset

View table

Table 5. Comparison of experimental results for different detection models on the ILS-MB dataset

Method	$m_{m A P_{50}}$	Accuracy rate of defect
Method	$m_{m A P_{50}}$	Baiban	Quesun	Yinbujun	Zhanyin	Xiuji	Huahen
Faster-RCNN （VGG16）	49.6	29.8	34.6	95.1	37.7	49.6	54.2
Faster-RCNN （ResNet50）	62.6	53.7	34.6	95.6	61.6	73.5	28.6
SSD （SSD300）	47.5	27.7	47.3	98.4	34.2	72.6	7.8
RetinaNet （ResNet50）	67.4	53.9	66.2	95.7	58.8	68.0	61.4
YOLO v3	61.4	40.1	72.5	93.6	67.5	56.3	38.2
YOLO v3tiny	72.2	46.8	70.5	99.5	78.0	84.8	53.6
YOLO v4	65.7	41.5	72.8	94.5	73.1	66.4	45.8
YOLO v5n	70.8	43.8	64.2	99.5	73.2	87.7	56.2
YOLO v5s	74.2	42.5	70.8	95.3	81.6	88.7	66.0
YOLO v5m	74.2	40.3	73.2	99.5	76.1	88.4	67.5
YOLO v5l	73.8	45.2	69.8	99.3	73.3	90.0	65.1
YOLO v7tiny	53.9	36.2	55.6	99.3	43.6	65.3	23.4
YOLOX	76.6	61.1	76.9	89.1	76.1	89.7	66.6
YOLO v8n	63.4	42.9	65.8	99.2	64.2	74	34.1
YOLO v8s	64.7	44	66.1	99.5	56.7	75.5	46.3
Metal-YOLO	78.3	46.6	72.1	99.5	87.6	96.3	67.5

Tools

Get Citation

Copy Citation Text

Bufan Zhang, Jinghu Yu, Xingfei Zhu, Zhaofei Sun, Yu Lu. Metal-YOLO Detection Algorithm for Defects in Coaxial Packaged Metal Base[J]. Laser & Optoelectronics Progress, 2024, 61(22): 2212003

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites