Confusion matrices comparing the networks’ predicted classification with the true classification for four different test mock datasets. The values shown are percentages. Top row: confusion matrices for the network trained with the single-label high S/N dataset without extinction (panel a) and for the network trained with the high S/N dataset with extinction (panel b). Bottom row: confusion matrices for the network trained with the low S/N dataset without extinction (panel c) and for the network trained with the multi-label dataset (panel d).