Table 2: Architecture of CNN-1 and CNN-2 in this study.

LayerInput sizeCNN-1Input sizeCNN-2

Conv_1256 × 256 × 13 × 3, 32256 × 256 × 13 × 3, 32
Dense-block_1256 × 256 × 32256 × 256 × 32
Max-pooling_1256 × 256 × 642 × 2, stride 2256 × 256 × 642 × 2, stride 2
Dense-block_2128 × 128 × 64128 × 128 × 64
Max-pooling_2128 × 128 × 1282 × 2, stride 2128 × 128 × 1282 × 2, stride 2
Dense-block_364 × 64 × 12864 × 64 × 128
Max-pooling_364 × 64 × 2562 × 2, stride 264 × 64 × 2562 × 2, stride 2
Dense-block_432 × 32 × 25632 × 32 × 256
Max-pooling_432 × 32 × 5122 × 2, stride 232 × 32 × 3842 × 2, stride 2
Dense-block_516 × 16 × 51216 × 16 × 384
Average-pooling16 × 16 × 102416 × 1616 × 16 × 51216 × 16
Fully connected layer102425122
Output22

Here, “conv” denotes convolutional layer. Number formats of CNN-1 and CNN-2 are all: convolution kernel size, number of convolution kernels.