Research Article

Project Gradient Descent Adversarial Attack against Multisource Remote Sensing Image Scene Classification

Table 6

Classification accuracy of the WHU-RS19 data set under the targeted attack.

ModelsOA of cleanOA of Xu2020OA of oursOA gap of Xu2020OA gap of ours

VGG-1691.03  0.8530.83  2.3026.61  1.74−60.20−64.42
GoogLeNet93.25  1.5447.80  1.4222.49  0.76−45.45−70.76
InceptionV391.72  0.5757.02  1.0231.45  0.75−34.70−60.27
ResNet-1891.52  1.2137.46  1.6515.91  1.14−54.05−75.60
ResNet-5093.90  0.4451.64  3.2024.42  2.19−42.25−69.48
ResNet-10192.14  0.8754.72  1.0227.58  1.65−37.43−64.56
DenseNet-12193.05  0.6959.63  1.6628.05  0.24−33.42−65.00
DenseNet-20194.26  0.5365.18  2.5032.53  0.64−29.08−61.73