Table 2.

The proposed CNN architecture used in our research (ReLU means rectified linear unit).

Layer Type	Number of Filters	Size of Feature Map (Height × Width × Channel)	Kernel (Filter) Size (Height × Width × Channel)	Number of Stride (Height × Width)	Number of Padding (Height × Width)
Image input layer		8 × 256 × 3
1st convolutional layer	64	8 × 244 × 64	1 × 13 × 3	1 × 1	0 × 0
Batch normalization		8 × 244 × 64
ReLU layer		8 × 244 × 64
2nd convolutional layer	64	8 × 232 × 64	1 × 13 × 64	1 × 1	0 × 0
Batch normalization		8 × 232 × 64
ReLU layer		8 × 232 × 64
Max pooling layer	1	8 × 116 × 64	1 × 2 × 64	1 × 2	0 × 0
3rd convolutional layer	128	8 × 104 × 128	1 × 13 × 64	1 × 1	0 × 0
Batch normalization		8 × 104 × 128
ReLU layer		8 × 104 × 128
4th convolutional layer	128	8 × 92 × 128	1 × 13 × 128	1 × 1	0 × 0
Batch normalization		8 × 92 × 128
ReLU layer		8 × 92 × 128
Max pooling layer	1	8 × 46 × 128	1 × 2 × 128	1 × 2	0 × 0
5th convolutional layer	256	8 × 36 × 256	1 × 11 × 128	1 × 1	0 × 0
Batch normalization		8 × 36 × 256
ReLU layer		8 × 36 × 256
6th convolutional layer	256	8 × 26 ×256	1 × 11 × 256	1 × 1	0 × 0
Batch normalization		8 × 26 × 256
ReLU layer		8 × 26 × 256
Max pooling layer	1	8 × 13 × 256	1 × 2 × 256	1 × 2	0 × 0
7th convolutional layer	512	6 × 11 × 512	3 × 3 × 256	1 × 1	0 × 0
Batch normalization		6 × 11 × 512
ReLU layer		6 × 11 × 512
8th convolutional layer	512	4 × 9 × 512	3 × 3 × 512	1 × 1	0 × 0
Batch normalization		4 × 9 × 512
ReLU layer		4 × 9 × 512
Max pooling layer	1	4 × 5 × 512	1 × 2 × 512	1 × 2	0 × 1
1st fully connected layer		4096
Batch normalization		4096
ReLU layer		4096
2nd fully connected layer		4096
Batch normalization		4096
ReLU layer		4096
3rd fully connected layer		# of classes
Softmax layer		# of classes
Classification layer (output layer)		# of classes