target and input matrix in Neural network (ANN)

Hi. I'm trying to learn Artificial Neural Network in Matlab and i want to do a simple task of recognizing if an image is a flower or not. So, i have taken 2 pictures of flowers and extracted 8 features from each of them which means i have 16 features in total. I'm using GUI instead of matlab commands to train my network. My question is how do i define the input and target matrix? in=[fa1, fb1 ; fa2, fb2 ;fa3, fb3 ;fa4, fb4 ;fa5, fb5 ;fa6, fb6 ;fa7, fb7 ;fa8, fb8]

where fa1 is the first feature of the first sample image and so on and fb1 is the first feature of the second sample image and so on

is this the correct way to define the input matrix? or do i use this matrix

in=[fa1, fa2, fa3, fa4 ,fa5, fa6 ,fa7, fa8 ;fb1, fb2 ,fb3, fb4 ,fb5, fb6 ,fb7, fb8]

Also, how do i define the target matrix? what will be the size of the target matrix? should it constitute of only 0s and 1s?

And lastly, once i've trained my network, how will i use it to determine if an unknown image is a flower or not?

Thank you

Answer 1

[Greg Heath]

For good generalization to unseen data:

For each class the number of samples should exceed the number of input features by a large factor. The input matrix for N I-dimensional inputs has the shape

 [ I N ] = size(input)% N >> I

Similarly, for the {0,1} c-class target matrix with columns from eye(c)

 [ c N ] = size(target)% N >> c

 The relationships between the class indices (1:c) and the target matrix are

 target           = ind2vec(trueclassindices)
 trueclassindices =  vec2ind(target)

 Similarly, the relationships between the classifier output matrix and the assigned class indices are

 assignedclassindices =  vec2ind(outputt)

 err = (assignedclassindices ~= trueclassindices)% (0,1) vector

 Nerr = sum(err)

 PctErr = 100*Nerr/N

Individual class errors are obtained by taking the class indices into account.

Hope this helps.

Thank you for formally accepting my answer

Greg

[The Zed]

You are a life saver Hats of!

Answer 2

[farheen asdf]

I have finally trained my neural network and the results are good (87% accurate). That being said i'm still a little confused as to how it can be used practically. For example, in my case it takes the network several tries to get to 87% accuracy. Sometimes its accuracy is as bad as 26%. How can i make sure that my network remembers what it has learned when it gets to 87% accuracy? Second, i was wondering if i could use this network to find the class of an unknown image which i select at runtime. I've used indexing method to separate the training, validation and test data so that the network tests only the images i want it to. Thanks in advance. Have a nice day :)