Table 2 Performance comparison of different DCNN models on DOTA data sets

VGG16

0.657

0.405

13.5 ms/1.4 ms/14.9 ms

321.8 M

VGG19

0.66

0.41

15.2 ms/1.4 ms/16.6 ms

364.2 M

InceptionV3

0.669

0.416

9.8 ms/1.3 ms/11.0 ms

394.5 M

InceptionV4

0.644

0.389

18.2 ms/1.2 ms/19.5 ms

543.5 M

ResNet50

0.677

0.424

11.8 ms/1.3 ms/13.1 ms

406.4 M

ResNet101

0.631

0.384

16.3 ms/1.4 ms/17.7 ms

558.7 M

ResNeXt50

0.677

0.424

20.7 ms/1.4 ms/22.2 ms

402.2 M

ResNeXt101

0.653

0.401

68.8 ms/1.3 ms/70.1 ms

558.7 M

SqueezeNet

0.651

0.396

7.3 ms/1.3 ms/8.6 ms

218.5 M

ShuffleNetV2

0.629

0.374

4.4 ms/1.3 ms/5.7 ms

217.8 M

Darknet53

0.68

0.432

15.3 ms/1.2 ms/16.5 ms

533.2 M

MobilenetV2

0.648

0.396

4.7 ms/1.3 ms/6.1 ms

218.5 M

MobileNetV3

0.658

0.399

5.9 ms/1.4 ms/7.3 ms

218.4 M

SE-ResNet50

0.672

0.42

12.5 ms/1.3 ms/13.8 ms

426.5 M

SK-ResNet50

0.663

0.411

13.0 ms/1.3 ms/14.3 ms

260.7 M

CSPDarknet53

0.705

0.45

14.1 ms/1.3 ms/15.4 ms

421.4 M

EfficientB0

0.646

0.394

7.9 ms/1.3 ms/9.2 ms

241.7 M

EfficientB1

0.646

0.392

10.0 ms/1.4 ms/11.4 ms

261.9 M

GhostNet

0.594

0.35

5.5 ms/1.4 ms/6.8 ms

229.9 M

Res2Net50

0.661

0.41

13.4 ms/1.4 ms/14.7 ms

407.6 M