Adding a sparseness constraint to the deep convolutional autoencoder