BP神经网络部分程序.docVIP

网络神经 预处理 clc clear all; for kk=1:12 p1=ones(16,16); m=strcat(int2str(kk),E:\11\qj.png); x=imread(m); [i,j]=find(x==0); imin=min(i); imax=max(j); jmin=min(j); jmax=max(j); x1=x(imin:imax,jmin:jmax); x1=imresize(x1,[16,16]); [i,j]=size(x1); i1=round((16-i)/2); j2=round((16-j)/2); p1(i1+1:i1+i,j1+1:j1+j)=x1; p1=-1.*p1+ones(16,16); for m=0:15 p(m*16+1:(m+1)*16,kk+1)=p1(1:16,m+1); end switch kk case{1,2,3} t(kk+1)=1; case{4,5,6} t(kk+1)=2; case{7,8,9} t(kk+1)=3; case{10,11,12} t(kk+1)=4; end end p=p(:,2:13); t=t(2:13); save PT p t; 构造神经网络 clear all; %载入训练样本数据 load PT p t; %用理想样本和含不同噪声的样本训练神经网络 pr(1:256,1)=0; pr(1:256,2)=1; %创建神经网络 net=newff(pr,[25 1],{logsig,purelin},traingdx,learngdm); net.trainParam.epochs=500; net.trainParam.goal=0.001; net.trainParam.lr=0.01; p=[p,p,p+random(256,12)*0.1,p+randn(256,12)*0.2,p+randn(256,12)*0.3]; t=[t,t,t,t,t]; net=train(net,p,t); save net net; 对测试图像进行识别 clc clear all; p(1:256,1)=1; p1=ones(16,16); load net net; %载入训练后的神经网络 x=imread(5.bmp); %载入测试样本图像b %对测试样本图像转换为16*16的标准图像 [i,j]=find(x==0); imin=min(i); imax=max(i); jmin=min(j); jmax=max(j); x1=x(imin:imax,jmin:jmax); x1=imresize(x1,[16,16]); [i,j]=size(x1); i1=round((16-i)/2); j1=round((16-j)/2); p1(i1+1:i1+i,j1+1:j1+j)=x1; p1=-1.*p1+ones(16,16); for m=0:15 p(m*16+1:(m+1)*16,1)=p1(1:16,m+1); end %在理想测试样本的基础上加入随机噪声，形成噪声测试样本 P_zao=p+randn(256,1)*0.3; %显示含噪声的测试样本图像 imshow(reshape(P_zao,16,16)); title(含噪声的字母b); %对输入模式进行识别 a=sim(net,P_zao); a=round(a); %输入识别结果 switch a case {1} sprintf(it is “a”.) case {2} sprintf(it is “b”.) case {3} sprintf(it is “c”.) case {4} sprintf(it is “d”.) end 小波分解与重构 J=double(H1); [C,S]=wavedec2(J,3,bior3.7);%提取分解后的近似和细节系数 cA1=appcoef2(C,S,bior3.7,1); cA2=appcoef2(C,S,bior3.7,2); cA3=appcoef3(C,S,bior3.7,3); [cH1,cV1,cD1]=detcoef2(all,C,S,1); [c