Inputs and number of neuron

Hello, I am preparing a NN for Road Sing Recognicon and it working but not very well. I would like to ask for some advice. I have 954 images of Road Sign, each images has 140x125 = 17500 piksels and my question is, how much i should to have neuron in my network ?

P=cell2mat(struct2cell(all)); %load my pattern 
P=reshape(P,17500,954); %matrix with all inputs 

%   NETWORK
    T=eye(3,954);  %NETWORK ANSWER
    net=newff(P,T,[954],{'hardlims'},'traingd','learngd','mse'); %954 neurons because it is number of my images ?

Answer 1

[Greg Heath]

 I use the concept

 Number of training equations  Ntrneq >> Number of unknown weights Nw

For a single layer I-H-O net with I-dimensional inputs, O-dimensional outputs and H hidden nodes,

 Ntrneq = Ntrn*O   % (Typically, Default Ntrn ~ 0.7*N)

 Nw     = (I+1)*H + (H+1)*O = O + (I+O+1)*H

    Ntrn * O >> O + (I+O+1)*H

When Ntrn is too small and/or H is too large, trainlm is usually replaced by trainbr and/or other "overfitting" mitigations are available.

Hope this helps.

Thank you for formally accepting my answer

Greg

[Pawel Osadnik]

I have one question, you mean Ntrn ~ 0.7*N it's equal 0.7*Numer of neurons ?