Evasion Attacks on Deep Learning-Based Helicopter Recognition Systems

Table 1

Model architecture.

Layer type	Shape
Conv. with ReLU	(224, 224, 64)
Max pooling	(112, 112, 64)
Conv. with ReLU	(112, 112, 128)
Max pooling	(2, 2)
Conv. with ReLU	(56, 56, 128)
Max pooling	(2, 2)
Conv. with ReLU	(28, 28, 128)
Max pooling	(2, 2)
FC with ReLU	(100,352)
FC with ReLU	(1,024)
FC with ReLU	(256)
FC with ReLU	(256)
FC with ReLU	(64)
FC with ReLU	(16)
Softmax	(2)

Conv. means convolutional layer. FC indicates the fully connected layer.