Table 1.

Specific parameters of the network structure of the hybrid CNN speckle noise removal model.

Subnet type	Layers	Type	Number of input and output feature maps	Convolution kernel size and stride
Noise	1	Conv + ReLU	1⟶64	3 × 3,1
	2–16	Conv + BN + ReLU	64⟶64	3 × 3,1
	17	Conv	64⟶1	3 × 3,1
Structure	1	Conv + ReLU	1⟶64	3 × 3,2
	2–5	Conv + ReLU	64⟶64	3 × 3,2
	6–9	Deconv + ReLU	64(+64)⟶64	3 × 3,2
	10	Deconv	64⟶1	3 × 3,2
Output	1	Conv + ReLU	1 + 1⟶64	3 × 3,1
	2–6	Conv + ReLU	64⟶64	3 × 3,1
	7	Conv	64⟶1	3 × 3,1