Detection and Segmentation of Structured Light Stripe in Weld Image

Fig. 6. Prediction results for each branch. (a)(b) Original images; (c)(d) results of Branch-dec; (e)(f) results of Branch-seg; (g)(h) total results; (i)(j) ground truth

Download full size

Fig. 7. Recognition results of different networks. (a)(b)(c)(d) Original images; (e)(f)(g)(h) test results of DT-Net; (i)(j)(k)(l) test results of Tiny DT-Net; (m)(n)(o)(p) test results of DT-ENet; (q)(r)(s)(t) test results of DT-MobileNet_UNet

Download full size

Table 1. Model comparison
View table
Table 1. Model comparison
Model Parameter FLOPs
U-Net 1.32×10⁷ 9.68×10¹¹
Modified U-Net 1.50×10⁶ 2.36×10¹⁰

Table 2. Prediction speed and accuracy of each branch
View table
Table 2. Prediction speed and accuracy of each branch
Model I_cls0 /% I_cls1 /% Mean IOU /% AP /% Time /ms Frame rate /(frame·s^-1)
Branch-seg 99.74 87.81 93.78 —— 40 25
Branch-dec —— —— —— 97.40 17 59
Total-result 99.74 87.81 93.78 97.40 42 24

Table 3. Performance comparison of different loss functions
View table
Table 3. Performance comparison of different loss functions
Loss I_cls0 /% I_cls1 /% Mean IOU /%
L_bce(Y^pred,Y^gt) 99.72 86.74 93.23
L_{bce_Dice}(Y^pred,Y^gt) 99.74 87.81 93.78

Table 4. Performance comparison of different networks

View table

Table 4. Performance comparison of different networks

Model	I_cls0 /%	I_cls1 /%	Mean IOU /%	AP /%	Time /ms	Frame rate /(frame·s^-1)
DT-Net	99.75	88.01	93.88	97.91	248	4
Tiny DT-Net	99.74	87.81	93.78	97.40	42	24
DT-ENet	99.72	86.73	93.22	97.40	135	7
DT-MobileNet_UNet	99.71	86.28	92.99	95.56	74	13

Tools

Get Citation

Copy Citation Text

Shikuan Zhang, Qingxiao Wu, Zhiyuan Lin. Detection and Segmentation of Structured Light Stripe in Weld Image[J]. Acta Optica Sinica, 2021, 41(5): 0515002

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites