CNN-Enabled Visibility Enhancement Framework for Vessel Detection under Haze Environment

Table 1

Network architecture of F-FFM.

Input	Layers	Type	Channels	Filter	Stride	Output	Size
	1	Convolutional	64		1	↓1
↓1	3	Convolutional	64		1	↓2
↓2		Max pooling	64		2	↓3
↓3	1	Convolutional	128		1	↓4
↓4	3	Convolutional	128		1	↓5
↓5		Max pooling	128		2	↓6
↓6	1	Convolutional	256		1	↓7
↓7	5	Convolutional	256		1	↓8
↓8		Bilinear interpolation				↓9
↓9↓4		Skip connection	128		1	↓10
↓10	1	Convolutional	128		1	↓11
↓11	3	Convolutional	128		1	↓12
↓12		Bilinear interpolation				↓13
↓13↓1		Skip connection	64		1	↓14
↓14	1	Convolutional	64		1	↓15
↓15	3	Convolutional	64		1	↓16
↓16		Global residual	3		1