SAR Ship Detection Based on Convolutional Neural Network with Deep Multiscale Feature Fusion

Fig. 8. Comparison of detection results of DMFFCNN when λ=0 and λ=3. (a) Detection results of SAR surface ships in scene interference when λ=0 ; (b) detection results of SAR surface ships in scene interference when λ=3; (c) detection results of SAR inshore ships in scene interference when λ=0; (d) detection results of SAR inshore ships in scene interference when λ=3

Download full size

Table 1. Comparison of experimental results
View table
Table 1. Comparison of experimental results
Method Backbone Input size AP /% Speed /(frame·s^-1)
SSD^[12] VGG16 300 76.1 45
FSSD^[15] VGG16 300 83.2 36
RFBNet^[18] VGG16 300 85.2 39
DMFFCNN(λ=0) Darknet-53 320 91.2 29

Table 2. Comparison of experimental results
View table
Table 2. Comparison of experimental results
Method Backbone Input size AP /% Speed /(frame·s^-1)
Feature fusion-1 Darknet-53 320 86.5 35
Feature fusion-2 Darknet-53 320 91.2 29

Table 3. Comparison of experimental results for different λ
View table
Table 3. Comparison of experimental results for different λ
λ AP /%
0 91.2
1 92.1
2 92.0
3 92.8
4 92.0
5 91.8

Tools

Get Citation

Copy Citation Text

Long Yang, Juan Su, Hua Huang, Xiang Li. SAR Ship Detection Based on Convolutional Neural Network with Deep Multiscale Feature Fusion[J]. Acta Optica Sinica, 2020, 40(2): 0215002

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites