[Retracted] Hybrid Depth-Separable Residual Networks for Hyperspectral Image Classification

Table 1

The architecture of the proposed network.

Layer	Output	Convolution kernel size	Convolution kernel number
Input	11, 11, 30, 1
C3D_1	9, 9, 24, 32	(3, 3, 7)	32
C3D_2	9, 9, 24, 32	(3, 3, 3)	32
C3D_3	9, 9, 24, 32	(3, 3, 3)	32
C3D_4	9, 9, 24, 32	(3, 3, 3)	32
C3D_5	9, 9, 24, 32	(3, 3, 7)	32
C3D	9, 9, 24, 32	(1, 1, 1)	32
C3D_6	7, 7, 1, 64	(3, 3, 24)	64
Reshape	7, 7, 64
C2D_1	5, 5, 128	(3, 3)	128
Sep_C2D_2	5, 5, 128	(3, 3)	128
Sep_C2D_3	5, 5, 128	(3, 3)	128
Sep_C2D_4	5, 5, 128	(3, 3)	128
Sep_C2D_5	5, 5, 128	(3, 3)	128
C2D	5, 5, 128	(1, 1)	128
Average pooling	1, 1, 128
FC	16
Output	Categories of land cover