Is it better to make neural network to have hierchical output?

Question

i'm quite new to neural network and i recently built neural network for number classification in vehicle license plate. It has 3 layers: 1 input layer for 16*24(382 neurons) number image with 150 dpi , 1 hidden layer(199 neurons) with sigmoid activation function, 1 softmax output layer(10 neurons) for each number 0 to 9.

I'm trying to expand my neural network to also classify letters in license plate. But i'm worried if i just simply add more classes into output, for example add 10 letters into classification so total 20 classes, it would be hard for neural network to separate feature from each class. And also, i think it might cause problem when input is one of number and neural network wrongly classifies as one of letter with biggest probability, even though sum of probabilities of all number output exceeds that.

So i wonder if it is possible to build hierchical neural network in following manner:

There are 3 neural networks: 'Item', 'Number', 'Letter'

1. 'Item' neural network classifies whether input is numbers or letters.

2. If 'Item' neural network classifies input as numbers(letters), then input goes through 'Number'('Letter') neural network.

3. Return final output from Number(Letter) neural network.

And learning mechanism for each network is below:

1. 'Item' neural network learns all images of numbers and letters. So there are 2 output.
2. 'Number'('Letter') neural network learns images of only numbers(letter).

Which method should i pick to have better classification? Just simply add 10 more classes or build hierchical neural networks with method above?

Answer 1

Item network should be recogniting regional register number standards and typical places of numbers on those standards. Otherwise your 'Item' phase has equal task than the whole system, because 'Letter' and 'Number' do not differ in plates so much you could right away choose between the two without first evaluating the exact object on each case.

Side note: correct me if that is not true on your local plate system

Answer 2

Just use one network with a larger Softmax output layer and more hidden units. If you have enough training data, it will work just fine. In fact it could emulate the architecture you propose.

Answer 3

I agree with the above answer. If you want to research this more in-depth look at this paper: https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/42241.pdf