Ship Detection in SAR Image Based on Multiple Connected Features Pyramid Network

Fig. 8. Comparison of experimental results between the proposed method and SSD algorithm. (a) Blurred background; (b) near shore and port area; (c) small ships near shore; (d) small ships on the sea

Download full size

Table 1. Experimental results of different parameters for focal loss
View table
Table 1. Experimental results of different parameters for focal loss
α_t γ AP /%
0.25 1 91.58
0.25 2 91.98
0.50 2 91.71
0.25 3 91.68
0.50 3 91.26
0.25 5 90.93

Table 2. Results of ablation experiments
View table
Table 2. Results of ablation experiments
VGG16 I-VGGNet FPN MCFPN SSD loss Ours loss AP /%
√ √ 88.34
√ √ 91.98
√ √ 92.70
√ √ √ 92.74
√ √ √ 93.60
√ √ √ 94.79

Table 3. Performance comparison with other methods on SAR ship dataset
View table
Table 3. Performance comparison with other methods on SAR ship dataset
Model Backbone AP /% Speed /(frame·s^-1)
SSD-300^[11] VGG16 88.34 43
SSD-512^[11] VGG16 89.43 17
Faster R-CNN^[8] VGG16 88.26 7
RetinaNet^[16] ResNet50 91.36 41
DSSD^[14] ResNet50 92.30 17
Proposed method I-VGGNet 94.79 22

Tools

Get Citation

Copy Citation Text

Zhijing Xu, Hai Huang. Ship Detection in SAR Image Based on Multiple Connected Features Pyramid Network[J]. Laser & Optoelectronics Progress, 2021, 58(8): 0811002

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites