Skip to main content
. 2021 Jan 3;21(1):268. doi: 10.3390/s21010268

Figure 14.

Figure 14

Res BCDU-Net architecture. The contraction path consists of Res blocks and a max-pooling layer. Such the U-Net, in each downsampling of encoding path, feature channels are doubled (64 to 128 to 256 to 512). In the last layer of the contracting path, we used 3 convolutional blocks with 2 dense connections. As seen, in the expansion path, the output of each batch normalized is given to a BConvLSTM layer.