Table 5.

The network architecture of our MDD-Net model.

Layer	Output Shape	Filter (Kernel size, Stride Size, Number)
Convolution2D	(None,60,50,24)	$3\times 3,1\times 1,24$
AveragePooling2D	(None,30,25,24)	$3\times 3,2\times 2$
Dense block 1	(None,30,25,36)	$\left[\begin{array}{ll}3\times 3,1\times 1,48& \mathrm{conv}\\ 3\times 3,1\times 1,12& \mathrm{conv}\end{array}\right]\times 3$
Transition block 1	(None,15,13,18)	$\left[\begin{array}{cc}3\times 3,1\times 1,30\hfill & \hfill \mathrm{conv}\\ 3\times 3,2\times 2\hfill & \hfill \mathrm{pooling}\end{array}\right]$
Dense block 2	(None,15,13,30)	$\left[\begin{array}{ll}3\times 3,2\times 2,48& \mathrm{conv}\\ 3\times 3,2\times 2,12& \mathrm{conv}\end{array}\right]\times 3$
Transition block 2	(None,8,7,15)	$\left[\begin{array}{cc}3\times 3,1\times 1,33\hfill & \hfill \mathrm{conv}\\ 3\times 3,2\times 2\hfill & \hfill \mathrm{pooling}\end{array}\right]$
Dense block 3	(None,8,7,27)	$\left[\begin{array}{ll}3\times 3,2\times 2,48& \mathrm{conv}\\ 3\times 3,2\times 2,12& \mathrm{conv}\end{array}\right]\times 3$
Maxpooling2D	(None,4,4,27)	$3\times 3,2\times 2$
Conv block 1	(None,30,25,24)	$\left[\begin{array}{ll}3\times 3,1\times 1,96& \mathrm{conv}\\ 3\times 3,1\times 1,24& \mathrm{conv}\end{array}\right]$
Concatenation 1	(None,30,25,96)	None
Maxpooling2D	(None,15,13,96)	$3\times 3,2\times 2$
Conv block 2	(None,15,13,12)	$\left[3\times 3,1\times 1,12 \mathrm{conv}\right]$
Conv block 3	(None,15,13,24)	$\left[\begin{array}{ll}3\times 3,1\times 1,96& \mathrm{conv}\\ 3\times 3,1\times 1,24& \mathrm{conv}\end{array}\right]$
Concatenation 2	(None,15,13,108)	None
Maxpooling2D	(None,8,7,108)	$3\times 3,2\times 2$
Conv block 4	(None,8,7,12)	$\left[3\times 3,1\times 1,12 \mathrm{conv}\right]$
Conv block 5	(None,8,7,24)	$\left[\begin{array}{ll}3\times 3,1\times 1,96& \mathrm{conv}\\ 3\times 3,1\times 1,24& \mathrm{conv}\end{array}\right]$
Concatenation 3	(None,8,7,108)	None
Maxpooling2D	(None,4,4,108)	$3\times 3,2\times 2$
Concatenation 4	(None,4,4,135)	None
Maxpooling2D	(None,2,2,135)	$3\times 3,2\times 2$
GlobalMaxPooling2D	(None,135)	None
Softmax	4	None