基于超像素和elm的图像分类方法源码

%利用超像素的特征来训练分类器。一块一块地输入分类器分类 输入超像素块数来得到分类结果 clc;clear; % no_classes = 16; no_classes = 9; no_train = round(1000); load pu1200 pulabel1200=labels; load('D:\ccpr论文程序\dataset\PaviaU.mat'); load('D:\ccpr论文程序\dataset\PaviaU_gt.mat'); load Indian_pines_gt %% img=paviaU; indian_pines_gt=paviaU_gt; indian_pines_corrected=paviaU; %% Resultfinal=[]; OAmax=0; CAmean=zeros(no_classes,1); for expi=1:1 tic [supermatrix,sup_img,indexescell]=generatemeanmatrix(indian_pines_corrected,pulabel1200); toc img=supermatrix; %选取训练集与测试集 [no_lines, no_rows, no_bands] = size(img); img = ToVector(img); img = img'; %每个样本选取的个数 % RandSampled_Num = round([300 300 300 300 300 300 300 300 300]);%pavia U RandSampled_Num = round([200 200 200 200 200 200 200 200 200]); Nonzero_map = zeros(no_lines,no_rows); Nonzero_index = find(indian_pines_gt ~= 0); Nonzero_map(Nonzero_index)=1; Train_Label = []; Train_index = []; for ii = 1: no_classes index_ii = find(indian_pines_gt == ii); class_ii = ones(length(index_ii),1)* ii; Train_Label = [Train_Label class_ii']; Train_index = [Train_index index_ii']; end trainall = zeros(2,length(Train_index)); trainall(1,:) = Train_index; trainall(2,:) = Train_Label; indexes =[]; for i = 1: no_classes W_Class_Index = find(Train_Label == i); Random_num = randperm(length(W_Class_Index)); Random_Index = W_Class_Index(Random_num); Tr_Index = Random_Index(1:RandSampled_Num(i)); indexes = [indexes Tr_Index]; end indexes = indexes'; train_SL = trainall(:,indexes); train_samples = img(:,train_SL(1,:))'; train_labels= train_SL(2,:)'; %选出部分分类结果作为训练样本 %1027个训练样本 % the remaining used for test test_SL = trainall; test_SL(:,indexes) = []; test_samples = img(:,test_SL(1,:))'; test_labels = test_SL(2,:)'; % Normalizing Training and original img [train_samples,M,m] = scale_func(train_samples); [img ] = scale_func(img',M,m); [sup_img ] = scale_func(sup_img',M,m); %elm------------------------------------------- train_samples=train_samples'; train_labels=train_labels'; img=img'; sup_img=sup_img'; tic [IW,B,LW,TF,TYPE] = elmtrain(train_samples,train_labels,2050,'sig',1);%330,'sig',1); Result=elmpredict(sup_img,IW,B,LW,TF,TYPE);%%ELM训练与测试过程 toc temp=zeros(1,no_lines*no_rows); for i=1:length(Result) temp(indexescell{1,i})=Result(i); end Result=temp;%%将每个特征向量对应的标号赋予对应超像素 Result = Result'; %elm------------------------------------ ResultTest = Result(test_SL(1,:),:); final=find(ResultTest==test_labels); GroudTest = double(test_labels(:,1)); ResultTest = Result(test_SL(1,:),:); [OA(expi),AA(expi),kappa(expi),CA(:,expi)]=confusion(GroudTest,ResultTest) if OA(expi)>OAmax OAmax=OA(expi); Resultfinal=Result;%5/22 end CAmean=CA(:,expi)+CAmean; end OA_mean=mean(OA) AA_mean=mean(AA) kappa_mean=mean(kappa)%%看十次平均值 revised 14/5/9 CAmean=CAmean/1 classmap=reshape(Resultfinal,no_lines, no_rows); resultmap1=classmap.*Nonzero_map; imagesc(resultmap1); axis off; % A=getframe; % imshow(A.cdata) % imwrite(A.cdata,'C:\Users\Think\Desktop\实验图像\2.jpg')