Fig. 1.

Different CNN-based deep learning architectures - Convolution block: conv - BN - ReLU - conv - BN - ReLU - Dropout (if enabled); Residual block: conv - BN - ReLU - conv - BN - shortcut - BN – shortcut + BN – ReLU; Dense block: Convolution block1 - concatenate(Input, Convolution block1) - Convolution block2 - concatenate(Input, Convolution block1, 2) - Convolution block3 - concatenate(Input, Convolution block1, 2, 3) - Convolution block4 - concatenate(Input, Convolution block1, 2, 3, 4); Where, Conv: Convolution; BN: Batch Normalization; ReLU: Activation.