Figure 4.

The architecture of the convolutional neural network, which was used in this work. The output dimensions of the layers are given at the top right of each row. The connections of the last two fully connected layers were omitted for better readability. The architecture was partly adapted from Fawaz et al. [29].