遗传算法的图像处理

%%%%%%%%%%%%%%%%遗传算法在道路图像阈值分割中的应用%%%%%%%%%%%% function main() clear all close all clc global chrom oldpop fitness lchrom popsize cross_rate mutation_rate thresholdsum global maxgen m n fit gen threshold A B C oldpop1 popsize1 b b1 fitness1 threshold1 A=imread('1.jpg'); %读入道路图像 A=imresize(A,0.5); %利用imresize函数通过默认的最近邻插值将图像放大0.5倍 B=rgb2gray(A); %灰度化 C=imresize(B,0.2); %将读入的图像缩小到0.2倍 lchrom=10; %染色体长度 popsize=10; %种群大小 cross_rate=0.8; %交叉概率 mutation_rate=0.5; %变异概率 maxgen=100; %最大代数 [m,n]=size(C); initpop; %初始种群 for gen=1:maxgen generation; %遗传操作 end findthreshold_best; %图象分割结果 %%%%%%%%%%%%%%%%%%%输出进化各曲线%%%%%%%%%%% figure; gen=1:maxgen; plot(gen,fit(1,gen)); title('最佳适应度值进化曲线'); xlabel('代数'), ylabel('最佳适应度值') figure; plot(gen,threshold(1,gen)); title('每一代的最佳阈值变化曲线'); xlabel('代数'), ylabel('每一代的最佳阈值') %%%%%%%%%%%%%%%%%%%初始化种群%%%%%%%%%%%%%%%%%%%% function initpop() global lchrom oldpop popsize chrom C imshow(C); for i=1:popsize chrom=rand(1,lchrom); for j=1:lchrom if chrom(1,j)<0.5 chrom(1,j)=0; else chrom(1,j)=1; end end oldpop(i,1:lchrom)=chrom; %给每一个个体分配8位的染色体编码 end %%%%%%%%%%%%%%%%%产生新一代个体%%%%%%%%%%%%%%%%%%%%%% function generation() fitness_order; %计算适应度值及排序 select; %选择操作 crossover; %交叉 mutation; %变异 %%%%%%%%%%%%%%%%%计算适度值并且排序%%%%%%%%%%%%%%%%%%% function fitness_order() global lchrom oldpop fitness popsize chrom fit gen C m n fitness1 thresholdsum global lowsum higsum u1 u2 threshold gen oldpop1 popsize1 b1 b threshold1 if popsize>=5 popsize=ceil(popsize-0.03*gen); end if gen==75 %当进化到末期的时候调整种群规模和交叉、变异概率 cross_rate=0.3; %交叉概率 mutation_rate=0.3; %变异概率 end %如果不是第一代则将上一代操作后的种群根据此代的种群规模装入此代种群中 if gen>1 t=oldpop; j=popsize1; for i=1:popsize if j>=1 oldpop(i,:)=t(j,:); end j=j-1; end end %计算适度值并排序 for i=1:popsize lowsum=0; higsum=0; lownum=0; hignum=0; chrom=oldpop(i,:); c=0; for j=1:lchrom c=c+chrom(1,j)*(2^(lchrom-j)); end b(1,i)=c*255/(2^lchrom-1); %转化到灰度值 for x=1:m for y=1:n if C(x,y)<=b(1,i) lowsum=lowsum+double(C(x,y));%统计低于阈值的灰度值的总和 lownum=lownum+1; %统计低于阈值的灰度值的像素的总个数 else higsum=higsum+double(C(x,y));%统计高于阈值的灰度值的总和 hignum=hignum+1; %统计高于阈值的灰度值的像素的总个数 end end end if lownum~=0 u1=lowsum/lownum; %u1、u2为对应于两类的平均灰度值 else u1=0; end if hignum~=0 u2=higsum/hignum; else u2=0; end fitness(1,i)=lownum*hignum*(u1-u2)^2; %计算适度值 end if gen==1 %如果为第一代，从小往大排序 for i=1:popsize j=i+1; while j<=popsize if fitness(1,i)>fitness(1,j) tempf=fitness(1,i); tempc=oldpop(i,:); tempb=b(1,i); b(1,i)=b(1,j); b(1,j)=tempb; fitness(1,i)=fitness(1,j); oldpop(i,:)=oldpop(j,:); fitness(1,j)=tempf; oldpop(j,:)=tempc; end j=j+1; end end for i=1:popsize fitness1(1,i)=fitness(1,i); b1(1,i)=b(1,i); oldpop1(i,:)=oldpop(i,:); end popsize1=popsize; else %大于一代时进行如下从小到大排序 for i=1:popsize j=i+1; while j<=popsize if fitness(1,i)>fitness(1,j) tempf=fitness(1,i); tempc=oldpop(i,:); tempb=b(1,i); b(1,i)=b(1,j); b(1,j)=tempb; fitness(1,i)=fitness(1,j); oldpop(i,:)=oldpop(j,:); fitness(1,j)=tempf; oldpop(j,:)=tempc; end j=j+1; end end end %下边对上一代群体进行排序 for i=1:popsize1 j=i+1; while j<=popsize1 if fitness1(1,i)>fitness1(1,j) tempf=fitness1(1,i); tempc=oldpop1(i,:); tempb=b1(1,i); b1(1,i)=b1(1,j); b1(1,j)=tempb; fitness1(1,i)=fitness1(1,j); oldpop1(i,:)=oldpop1(j,:); fitness1(1,j)=tempf; oldpop1(j,:)=tempc; end j=j+1; end end %下边统计每一代中的最佳阈值和最佳适应度值 if gen==1 fit(1,gen)=fitness(1,popsize); threshold(1,gen)=b(1,popsize); thresholdsum=0; else if fitness(1,popsize)>fitness1(1,popsize1) threshold(1,gen)=b(1,popsize); %每一代中的最佳阈值 fit(1,gen)=fitness(1,popsize);%每一代中的最佳适应度 else threshold(1,gen)=b1(1,popsize1); fit(1,gen)=fitness1(1,popsize1); end end %%%%%%%%%%%%%%%%%%%精英选择%%%%%%%%%%%%%%%%%%%% function select() global fitness popsize oldpop temp popsize1 oldpop1 gen b b1 fitness1 %统计前一个群体中适应值比当前群体适应值大的个数 s=popsize1+1; for j=popsize1:-1:1 if fitness(1,popsize)<fitness1(1,j) s=j; end end for i=1:popsize temp(i,:)=oldpop(i,:); end if s~=popsize1+1 if gen<50 %小于50代用上一代中用适应度值大于当前代的个体随机代替当前代中的个体 for i=s:popsize1 p=rand; j=floor(p*popsize+1); temp(j,:)=oldpop1(i,:); b(1,j)=b1(1,i); fitness(1,j)=fitness1(1,i); end else if gen<100 %50~100代用上一代中用适应度值大于当前代的个体代替当前代中的最差个体 j=1; for i=s:popsize1 temp(j,:)=oldpop1(i,:); b(1,j)=b1(1,i); fitness(1,j)=fitness1(1,i); j=j+1; end else %大于100代用上一代中的优秀的一半代替当前代中的最差的一半，加快寻优 j=popsize1; for i=1:floor(popsize/2) temp(i,:)=oldpop1(j,:); b(1,i)=b1(1,j); fitness(1,i)=fitness1(1,j); j=j-1; end end end end %将当前代的各项数据保存 for i=1:popsize b1(1,i)=b(1,i); end for i=1:popsize fitness1(1,i)=fitness(1,i); end for i=1:popsize oldpop1(i,:)=temp(i,:); end popsize1=popsize; %%%%%%%%%%%%%交叉%%%%%%%%%%%%%%%%%%%%%%%%%% function crossover() global temp popsize cross_rate lchrom j=1; for i=1:popsize p=rand; if p<cross_rate parent(j,:)=temp(i,:); a(1,j)=i; j=j+1; end end j=j-1; if rem(j,2)~=0 j=j-1; end if j>=2 for k=1:2:j cutpoint=round(rand*(lchrom-1)); f=k; for i=1:cutpoint temp(a(1,f),i)=parent(f,i); temp(a(1,f+1),i)=parent(f+1,i); end for i=(cutpoint+1):lchrom temp(a(1,f),i)=parent(f+1,i); temp(a(1,f+1),i)=parent(f,i); end end end %%%%%%%%%%%%%%%%%%%%%%%%%变异%%%%%%%%%%%%%%%%%%%%% function mutation() global popsize lchrom mutation_rate temp newpop oldpop sum=lchrom*popsize; %总基因个数 mutnum=round(mutation_rate*sum); %发生变异的基因数目 for i=1:mutnum s=rem((round(rand*(sum-1))),lchrom)+1; %确定所在基因的位数 t=ceil((round(rand*(sum-1)))/lchrom); %确定变异的是哪个基因 if t<1 t=1; end if t>popsize t=popsize; end if s>lchrom s=lchrom; end