Shape recognition by neural network

Hello, I would like to ask which Neural Network should I choose. I want to recognition of 3 shape (Square,Circle,Triangle).

I learn my NN with 69 patterns (23 image of Square, 23 Circle etc). First network was "newp" and the second "newff". I have some answers but my network recognition e.g Triangle as Square and Circle or something like this. Any ideas ? Thank !

%    PATTERNS

load PatternsSquare
load PatternsCircle
load PatternsTriangle
%    SHAPE Square
P1=cell2mat(struct2cell(PatternsSquare));
PS=reshape(P1,40000,23);
%   SHAPE Circle
P2=cell2mat(struct2cell(PatternsCircle));
PO=reshape(P1,40000,23);
%   SHAPE Triangel
P3=cell2mat(struct2cell(PatternsTriangle));
PT=reshape(P1,40000,23);

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

NETWORK

P=[PS,PO,PT]; %MATRIX WITH ALL MY PATTERN 
T=eye(3,69);  %NETWORK ANSWER
    %net = newp(P,T,'hardlim','learnp');
    net=newff(P,T,[10 10],{'tansig' 'purelin'},'traingd');
    net.trainParam.epochs = 50;
    net.trainParam.goal=0.00001;
    net = train(net,P,T);
    Y=sim(net,P);   

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

DISRUPTION

   D=imread('T14.png'); % MY DISRUPTION PATTERN 
   D=im2bw(D);
   D=double(D);
   [m n] = size(D) ; 
   D = imresize(D, [200 200]); 
   [x y] = size(D); 
  PD=reshape(D,40000,1);
  YD=sim(net,PD);

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%  DISPLAY

figure(1)
 for i=1:1
    subplot(1,2,i) 
    imshow(reshape(PD(:,i),200,200))
    title('DISRUPTION')
    subplot(1,2,2*i)
    hintonw(YD(:,i))
    title('NEURON: 1-Square, 2-Circle, 3-Triangle')
 end

Answer 1

[Greg Heath]

 1.You have erroneouslyy defined PO and PT with P1
2. I think the argument in reshape should be (Pm,23,40000)
3. You have 2 hidden layers. One is sufficient.
4. I didn't look at Disruption and Display

Hope this helps.

Greg

[Pawel Osadnik]

After when I change argument in reshape matrix occur an error:

Error using network/train (line 340)
Inputs and targets have different numbers of samples.

Error in SSN (line 26)
    net = train(net,P,T);
My input data are 23 elements of 17.500 sampels (image piksel). And the data to chceck (DISRUPTION) are 1 of 17.500 and it's different. 
It doesn't work.