exampletr.m

数字识别基于MATLAB的数字识别系统的研究与实现

%NN1T,Training of BP network
%==============
%BP网络训练
%==============
figure('name','训练过程图示','numbertitle','off');

S(:,1)=nn1f('00.bmp');
S(:,2)=nn1f('10.bmp');
S(:,3)=nn1f('20.bmp');
S(:,4)=nn1f('30.bmp');
S(:,5)=nn1f('40.bmp');
S(:,6)=nn1f('50.bmp');
S(:,7)=nn1f('60.bmp');
S(:,8)=nn1f('70.bmp');
S(:,9)=nn1f('80.bmp');
S(:,10)=nn1f('90.bmp');

T=[0 1 2 3 4 5 6 7 8 9];    %训练样本的期望输出   1个输出神经元
 %  T=[ 0 0 0 0 0 0 0 0 0 1;
  %     0 0 0 1 1 1 2 2 2 0;
   %    0 1 2 0 1 2 0 1 2 0];

[R,Q]=size(S);[S2,Q]=size(T);

S1=str2num(S1);  %隐层神经元个数

%W1=rands(R,S1);
[W1,B1]=rands(S1,R);
[W2,B2]=rands(S2,S1);
A2=purelin(W2*tansig(W1*S,B1),B2);

%initialize the parameters
disp_freq=20;
max_epoch=str2num(max_epoch);
err_goal=str2num(err_goal);
lr=str2num(lr);  %学习步长
TP=[disp_freq max_epoch err_goal lr];
%training begins
[W1,B1,W2,B2,epochs,errors]=trainbp(W1,B1,'tansig',W2,B2,'purelin',S,T,TP);

%ploterr(errors);
out=purelin(W2*tansig(W1*S,B1),B2);  %实际输出

SSE=sumsqr(T-out);

fprintf('Trained network operates:');
if SSE < err_goal
   disp('Adequately.')
else
   disp('Inadequately.')
end