Table 2 The layer-wise summary of the single U-Net and the sub-classifiers used in the CEU-Net architecture. n x n is the patch size, w is the input spectral dimension, and m is the class size for the given dataset

0

Input Layer

N/A

(n,n,w)

1A

Conv2D_1

Kernel = (3,3), strides = (1,1)

0

(n,n,64)

1B

BatchNormalization_1

1A

(n,n,64)

1C

LeakyReLU_1

1B

(n,n,64)

1D

Dropout_1

0.2 Dropped

1C

(n,n,64)

2A

Conv2D_2

Kernel = (3,3), strides = (1,1)

1D

(n,n,128)

2B

BatchNormalization_2

2A

(n,n,128)

2C

LeakyReLU_2

2B

(n,n,128)

2D

Dropout_2

0.2 Dropped

2C

(n,n,128)

3A

Conv2D_3

Kernel = (3,3), strides = (1,1)

2D

(n,n,256)

3B

BatchNormalization_3

3A

(n,n,256)

3C

LeakyReLU_3

3B

(n,n,256)

3D

Dropout_3

0.2 Dropped

3C

(n,n,256)

4A

Conv2DTranspose_1

Kernel = (3,3), strides = (1,1)

3D

(n,n,256)

4B

BatchNormalization_4

4A

(n,n,256)

4C

LeakyReLU_4

4B

(n,n,256)

4D

Dropout_4

0.2 Dropped

4C

(n,n,256)

4E

Concatenate_1

2D + 4D

2D,4D

(n,n,384)

5A

Conv2DTranspose_2

Kernel = (3,3), strides = (1,1)

4E

(n,n,128)

5B

BatchNormalization_5

5A

(n,n,128)

5C

LeakyReLU_5

5B

(n,n,128)

5D

Dropout_5

0.2 Dropped

5C

(n,n,128)

5E

Concatenate_2

1D + 5D

1D, 5D

(n,n,192)

6A

Conv2DTranspose_3

Kernel = (3,3), strides = (n,n)

5E

(1,1,m)

6B

Reshape

6A

(1,m)

6C

PixelSoftmax

6B

(1,m)