The model architecture of the CNN. The epochal time-resolved difference images for a transient candidate are arranged as a 61 × 61 × 18 array as an input for the model; the example shown are the difference images for an echo location near Cas A with an extended structure seen propagating from left to right in the difference images. ReLU activation is used for the convolutional and dense layers, and softmax activation is used for the final layer to map class probabilities between 0 and 1.