Low-Altitude Sea Surface Infrared Object Detection Based on Unsupervised Domain Adaptation

Fig. 5. Comparison of results between supervised detection and unsupervised domain adaptation detection. (a) Supervised detection; (b) unsupervised domain adaptation detection

Download full size

Table 1. Influence of GRL in different positions on detection accuracy
View table
Table 1. Influence of GRL in different positions on detection accuracy
GRL position mAP /%
w/o GRL(YOLOv5s) 84.84
Feature pyramid 89.13
Feature pyramid+head 90.34

Table 2. Influence of different λ_MMD values on detection accuracy
View table
Table 2. Influence of different λ_MMD values on detection accuracy
λ_MMD mAP /%
0.5 89.56
1.0 92.18
2.0 91.98

Table 3. Influence of different training optimization algorithms on detection accuracy
View table
Table 3. Influence of different training optimization algorithms on detection accuracy
Optimization algorithm mAP /%
SGD synchronization 88.11
SGD asynchronization 90.84
AdamW synchronization 85.18
AdamW asynchronization 92.18

Table 4. Comparison of detection accuracy between source domain and target domain

View table

Table 4. Comparison of detection accuracy between source domain and target domain

Serial number	Training dataset	Test dataset	Detection accuracy /%
Serial number	Training dataset	Test dataset	Source domain	Target domain
1	Source domain	Source domain	94.32	90.31
2	Source domain	Target domain	81.94	84.84
3	Target domain	Target domain	87.33	86.64
4	Source domain+target domain	Source domain+target domain	89.26	89.56
5	Supervised domain adaptation	Target domain	95.85	92.44
6	Unsupervised domain adaptation	Target domain	95.82	92.18

Tools

Get Citation

Copy Citation Text

Zizhuang Song, Jiawei Yang, Dongfang Zhang, Shiqiang Wang, Yue Zhang. Low-Altitude Sea Surface Infrared Object Detection Based on Unsupervised Domain Adaptation[J]. Acta Optica Sinica, 2022, 42(4): 0415001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites