Figure 2.

Overview of the four convolutional neural network architectures used for binary classification of start and stop features extracted from wear and non-wear episodes. Before the first fully connected layer, the output data from the previous layer is flattened.