Table 3.

Structure of CNNs model used in our work.

Layer		Kernel Size	Filters Number	Feature Map Size	Stride	Padding	Iterations Number
Image Layer				$224\times 224\times 3$
Conv-1		$7\times 7$	64	$112\times 112$	2	$3\times 3$	1
Max pool		$3\times 3$	1	$56\times 56$	2	-	1
Conv-2	Conv-2-1	$1\times 1$	64	$56\times 56$	1	0	3
	Conv-2-2	$3\times 3$	64		1	1
	Conv-2-3	$1\times 1$	256		1	0
Conv-3	Conv-3-1	$1\times 1$	128	$28\times 28$	2	0	4
	Conv-3-2	$3\times 3$	128		1	1
	Conv-3-3	$1\times 1$	512		1	0
Conv-4	Conv-4-1	$1\times 1$	256	$14\times 14$	2	0	23
	Conv-4-2	$3\times 3$	256		1	1
	Conv-4-3	$1\times 1$	1024		1	0
Conv-5	Conv-5-1	$1\times 1$	512	$7\times 7$	2	0	3
	Conv-5-2	$3\times 3$	512		1	1
	Conv-5-3	$1\times 1$	2048		1	0
Average pool		$1\times 1$	1	$7\times 7$	1	0	1
Full Connect layer		10	-	-	-	-	1
Softmax		10	-	-	-	-	1