Defect detection in ceramic substrate based on improved YOLOV4

Feature map size	Anchor Size
$19 \times 19$	（108×121）；（20×86）；（113×94）
$38 \times 38$	（51×49）；（54×42）；（36×53）
$76 \times 76$	（26×24）；（24×18）；（22×31）
$152 \times 152$	（7×9）；（11×14）；（16×12）

Table 2. Results of defect detection

View table

View in Article

Table 2. Results of defect detection

Defect	Total	Correct detection	False detection	Missed detection	Precision/%	Recall/%
Stain	1 013	1 004	5	9	99.5	99.1
Foreign matter	22	22	0	0	100	100
Gold edge bulge	83	83	3	0	96.5	100
Damage	185	182	15	4	92.4	97.9
Ceramic gap	74	73	0	1	100	98.6
Total	1 377	1 364	23	14	98.3	99.0

Table 3. Defect detection results of different algorithms

View table

View in Article

Table 3. Defect detection results of different algorithms

Model	Size	Precision/%	Recall/%	Times/s
Faster R-CNN	$(600 \times 600)$	83.3	82.9	0.217
Efficientdet-B3	$(768 \times 768)$	81.3	80.5	0.256
YOLOV4	$(608 \times 608)$	86.9	90.1	0.182
YOLOV5	$(640 \times 640)$	87.7	89.8	0.176
YOLOX	$(640 \times 640)$	85.1	86.5	0.167
YOLOV4-CS	$(608 \times 608)$	98.3	99.0	0.202

Table 4. Results of different ablation experiments
View table
View in Article
Table 4. Results of different ablation experiments
Small scale branch CCNet GHMC Precision/% Recall/%
√ 93.6 93.3
√ 90.1 91.2
√ 91.9 95.0
√ √ 96.7 94.2
√ √ 96.5 97.5
√ √ 95.6 97.9
√ √ √ 98.3 99.0

Table 5. Results of different kinds of attention networks
View table
View in Article
Table 5. Results of different kinds of attention networks
Attentional mechanisms Precision/% Recall/% Time/s
SENet 96.4 95.1 0.201
CBAM 97.5 97.6 0.227
ECA-Net 97.1 97.4 0.222
CCNet 98.3 99.0 0.202

Table 6. Results of different kinds of prior box
View table
View in Article
Table 6. Results of different kinds of prior box
Prior box Precision/% Recall/%
YOLOV4 96.3 96.6
Ours 98.3 99.0

Tools

Get Citation

Copy Citation Text

Feng GUO, Qibing ZHU, Min HUANG, Xiaoxiang XU. Defect detection in ceramic substrate based on improved YOLOV4[J]. Optics and Precision Engineering, 2022, 30(13): 1631

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites