SiameseDenseU-Net-based Semantic Segmentation of Urban Remote Sensing Images

Table 1

Detailed parameters of SiameseDenseU-Net.

Layer name		Kernel number	Kernel size
Inconv		32	3 × 3
TOP image
Down	Conv0_1	32	3 × 3
	Conv0_2	32	3 × 3
Block0	Conv0_3	32	1 × 1
	Maxpool0	32	2 × 2
Down	Conv1_1	64	3 × 3
	Conv1_2	64	3 × 3
Block1	Conv1_3	64	1 × 1
	Maxpool1	64	2 × 2
Down	Conv2_1	128	3 × 3
	Conv2_2	128	3 × 3
Block2	Conv2_3	128	1 × 1
	Maxpool2	128	2 × 2
Down	Conv3_1	256	3 × 3
	Conv3_2	256	3 × 3
Block3	Conv3_3	256	1 × 1
	Maxpool3	256	2 × 2
Down	Conv4_1	256	3 × 3
	Conv4_2	256	3 × 3
Block4	Conv4_3	256	1 × 1
	Maxpool4	256	2 × 2
Up	TransposedConv0	256	2 × 2
	Conv5_1	256	1 × 1
Block0	Conv5_2	256	3 × 3
	Conv5_3	256	3 × 3
	Conv5_4	256	1 × 1
Up	TransposedConv1	256	2 × 2
	Conv6_1	128	1 × 1
Block1	Conv6_2	128	3 × 3
	Conv6_3	128	3 × 3
	Conv6_4	128	1 × 1
Up	TransposedConv2	128	2 × 2
	Conv7_1	64	1 × 1
Block2	Conv7_2	64	3 × 3
	Conv7_3	64	3 × 3
	Conv7_4	64	1 × 1
Up	TransposedConv3	64	2 × 2
	Conv8_1	32	1 × 1
Block3	Conv8_2	32	3 × 3
	Conv8_3	32	3 × 3
	Conv8_4	32	1 × 1
Up	TransposedConv4	32	2 × 2
	Conv9_1	32	1 × 1
Block4	Conv9_2	32	3 × 3
	Conv9_3	32	3 × 3
	Conv9_4	32	1 × 1
nDSM
Down	Conv0_1	32	3 × 3
	Conv0_2	32	3 × 3
Block0	Conv0_3	32	1 × 1
	Maxpool0	32	2 × 2
Down		64	3 × 3
		64	3 × 3
Block1		64	1 × 1
		64	2 × 2
Down	Conv2_1	128	3 × 3
	Conv2_2	128	3 × 3
Block2	Conv2_3	128	1 × 1
	Maxpool2	128	2 × 2
Down	Conv3_1	256	3 × 3
	Conv3_2	256	3 × 3
Block3	Conv3_3	256	1 × 1
	Maxpool3	256	2 × 2
Down	Conv4_1	256	3 × 3
	Conv4_2	256	3 × 3
Block4	Conv4_3	256	1 × 1
	Maxpool4	256	2 × 2
Up	TransposedConv0	256	2 × 2
	Conv5_1	256	1 × 1
Block0	Conv5_2	256	3 × 3
	Conv5_3	256	3 × 3
	Conv5_4	256	1 × 1
Up	TransposedConv1	256	2 × 2
	Conv6_1	128	1 × 1
Block1	Conv6_2	128	3 × 3
	Conv6_3	128	3 × 3
	Conv6_4	128	1 × 1
Up	TransposedConv2	128	2 × 2
	Conv7_1	64	1 × 1
Block2	Conv7_2	64	3 × 3
	Conv7_3	64	3 × 3
	Conv7_4	64	1 × 1
Up	TransposedConv3	64	2 × 2
	Conv8_1	32	1 × 1
Block3	Conv8_2	32	3 × 3
	Conv8_3	32	3 × 3
	Conv8_4	32	1 × 1
Up	TransposedConv4	32	2 × 2
	Conv9_1	32	1 × 1
Block4	Conv9_2	32	3 × 3
	Conv9_3	32	3 × 3
	Conv9_4	32	1 × 1
Outconv (concatenata TOP and nDSM feature)		6	3 × 3