matlab简单神经网络示例

一、模型训练

sample_in = []

sample_out = []
offset = 5


path='D:\demo3\l\';
Files = dir(fullfile(path,'*.png'));
LengthFiles = length(Files);
a_s_out = [1 0 0];
for i=1:LengthFiles
    Img = imread([path,Files(i).name]); 
    Img_grey = rgb2gray(Img);
    a_sample = [];
    for i = 1:offset:size(Img_grey,1)
        for j = 1:offset:size(Img_grey,2)
            a_sample(end+1) = Img_grey(i,j);
        end
    end
    sample_in = [sample_in;a_sample];
    sample_out = [sample_out;a_s_out];
    %plot(a_sample);
    %imagesc(Img2);
end
size(sample_in)


path='D:\demo3\r\';
Files = dir(fullfile(path,'*.png'));
LengthFiles = length(Files);
a_s_out = [0 0 1];
for i=1:LengthFiles
    Img = imread([path,Files(i).name]); 
    Img_grey = rgb2gray(Img);
    a_sample = [];
    for i = 1:offset:size(Img_grey,1)
        for j = 1:offset:size(Img_grey,2)
            a_sample(end+1) = Img_grey(i,j);
        end
    end
    sample_in = [sample_in;a_sample];
    sample_out = [sample_out;a_s_out];
    %plot(a_sample);
    %imagesc(Img2);
end


path='D:\demo3\m\';
Files = dir(fullfile(path,'*.png'));
LengthFiles = length(Files);
a_s_out = [0 1 0];
for i=1:LengthFiles
    Img = imread([path,Files(i).name]); 
    Img_grey = rgb2gray(Img);
    a_sample = [];
    for i = 1:offset:size(Img_grey,1)
        for j = 1:offset:size(Img_grey,2)
            a_sample(end+1) = Img_grey(i,j);
        end
    end
    sample_in = [sample_in;a_sample];
    sample_out = [sample_out;a_s_out];
    %plot(a_sample);
    %imagesc(Img2);
end




 p = sample_in';
 t = sample_out';%[1 0;0 1]';


% % %创建BP网络
 net = newff(minmax(p),[100 3],{'logsig' 'logsig'},'traingda');
% % 
% % %初始化BP网络，包括：权值和偏置的初值
  net = init(net);
% % 
% % %设置训练参数和训练BP网络
  net.trainParam.epochs = 200;
  net.trainParam.goal = 0.001;
  net.trainParam.show = 10;
% % %训练BP神经网络
net = train(net,p,t);


%保存网络
save('jtt_net.mat','net');


%自洽验证，
BPoutput=sim(net,p); 
str = 'BPoutput size is';
size(BPoutput)
figure;
imagesc(BPoutput);


order = [];
for j = 1:size(BPoutput,2)%very sample
    big_order = 1;
    big_val = BPoutput(1,j);
    for i = 2: size(BPoutput,1)
        if big_val < BPoutput(i,j)
            big_order = i;
        end
    end
    order(end+1) = big_order;
end

figure;plot(order,'*');

二、模型使用

load('jtt_net.mat');
sample_in = []
sample_out = []
offset = 5


path='D:\demo3\before\';
Files = dir(fullfile(path,'*.png'));
LengthFiles = length(Files);
for i=1:LengthFiles
    Img = imread([path,Files(i).name]); 
    Img_grey = rgb2gray(Img);
    a_sample = [];
    for i = 1:offset:size(Img_grey,1)
        for j = 1:offset:size(Img_grey,2)
            a_sample(end+1) = Img_grey(i,j);
        end
    end
    sample_in = [sample_in;a_sample];
end


p = sample_in';
size(p)
size(t)
BPoutput=sim(net,p); 


str = 'BPoutput size is';
size(BPoutput)
figure;
imagesc(BPoutput);


order = [];
for j = 1:size(BPoutput,2)%very sample
    big_order = 1;
    big_val = BPoutput(1,j);
    for i = 2: size(BPoutput,1)
        if big_val < BPoutput(i,j)
            big_order = i;
        end
    end
    order(end+1) = big_order;
end
figure;plot(order,'*');


%文件转移
for i=1:LengthFiles
    if order(i) == 1
        copyfile([path,Files(i).name],[path,'\left\',Files(i).name]);
    elseif (order(i)==2)
        copyfile([path,Files(i).name],[path,'\middle\',Files(i).name]);
    elseif (order(i)==3)
        copyfile([path,Files(i).name],[path,'\right\',Files(i).name]);
    end;        
end;