how to define target matrix for rbf neural classifier

i have 10 original images and 10 corrupted images. the image features are in 10 X 548 matrices. i want to classify the original images and corrupted images separately. so i should classify the images as '1' if original image and classify as '0' if corrupted image. 1. how to define the target matrix? 2. what values to use for goal and spread in newrb? 3. how to simulate the network?

Answer 1

[Greg Heath]

target = [ ones(1,10) zeros(1,10)]

MSE00 = mean(var(target',1))

MSEgoal = MSE00/100

Standardize inputs (ZSCORE or MAPSTD) and search for spreads about unity

spread(1) = 1

spread(i+1) = 0.5*spread(i)

and

spread(j+1) = 2*spread(j)

Hope this helps

Greg