Question

My project is to recognize ancient coins. I am using Matlab. I already have a feature file which contains both inputs and output. I have trained 3 types of coins using newff and net had been saved. For the three types of coins, I used 01, 10 and 11 as targets. Now I want to use that trained net for testing. I have test images too. I coded like this:

load net.mat;
load features.mat;
testInputs = Features';
out = sim(net,testInputs);
[dummy, I]=max(out);

Value of I is using to check the coin type. If I is 1 then type 1, if 2 then type 2 and if 3 type 3. Am I correct? I hard coded these 1,2,3 values because I gave targets as 01, 10 and 11.

if (I == 2)
    fprintf('Type1\n');
elseif (I == 1)
    fprintf('Type2\n');
elseif (I == 3)
    fprintf('Type3\n');
else
    fprintf('undefined\n');
end

Although now I input 3 types of test coin images, it either displays 1 or 2 for the value I. But not 3. Even when I am using the same set of images which are used for training, it also gives either 1 or 2 for the value I.

Can u please help me?

Answer 1

The second argument of max() will give you the index of the neuron with higher output. If you have only two neurons, which is the case if your targets are [0,1], [1,0] and [1,1] (note only two elements on every target) there will be no way to get a 3 out of that max(). You should try [0,0,1], [0,1,0] and [1,0,0].

On a side note, if you are using tansig as the activation function of the neurons, consider using -1 instead of 0 on the targets, so you can better exploit the non-linearity. Something like [-1,-1,1], [-1,1,-1], [1,-1,-1].