神经网络遥感图像分类matlab

clc clear all %% found colorset colorset = zeros(10,3); %色彩1 colorset(1,1) = 255; colorset(1,2) = 0; colorset(1,3) = 0; %色彩2 colorset(2,1) = 0; colorset(2,2) = 255; colorset(2,3) = 0; %色彩3 colorset(3,1) = 0; colorset(3,2) = 0; colorset(3,3) = 255; %色彩4 colorset(4,1) = 255; colorset(4,2) = 255; colorset(4,3) = 0; %色彩5 colorset(5,1) = 125; colorset(5,2) = 125; colorset(5,3) = 125; %色彩6 colorset(6,1) = 0; colorset(6,2) = 0; colorset(6,3) = 125; %色彩7 colorset(7,1) = 0; colorset(7,2) = 255; colorset(7,3) = 255; %色彩8 colorset(8,1) = 255; colorset(8,2) = 0; colorset(8,3) = 255; %色彩9 colorset(9,1) = 125; colorset(9,2) = 125; colorset(9,3) = 0; %色彩10 colorset(10,1) = 125; colorset(10,2) = 125; colorset(10,3) = 0; %... %色彩n %% 输入训练数据和待分类图像信息 % line = 200; % sample = 200; %% 读取整幅遥感图像 %% 调用ReadImage函数 flag1 = ReadImage(); %% 读取ROI数据 [FileName,PathName,FilterIndex] = uigetfile({'*.txt','Text Files (*.txt)'},'请选择训练样本'); if (FilterIndex==0) flag = 0; return; end path=fullfile(PathName,FileName); fin = fopen(path,'r'); tline = fgetl(fin); %% 略去注释行 i=1; while (tline(1) == ';') flag = strfind(tline,'; Number of ROIs'); if isempty(flag)==0 class_num = sscanf(tline,'; Number of ROIs: %d'); tline = fgetl(fin); continue; end flag = strfind(tline,'; ROI npts'); if isempty(flag)==0 num(i) = sscanf(tline,'; ROI npts: %d'); i = i + 1; tline = fgetl(fin); continue; end tline = fgetl(fin); end %% 生成纯数值文本 path = [path,'_new.txt']; fout = fopen(path ,'w'); fprintf(fout,'%s\n',tline); while ~feof(fin) tline = fgetl(fin); fprintf(fout,'%s\n',tline); end fclose(fout); fclose(fin); %% 调用Readtrain函数 flag2 = Readtrain(path,num); train_num = sum(num); %% 读入数据 %% %%载入训练数据 if flag2==0 % msgbox('请先选择训练样本数据！','错误','error','modal'); return; else load train.mat P = train_data'; T = train_label; end %% %%载入待分类图像数据 if flag1==0 % msgbox('请先选择待分类的遥感图像！','错误','error','modal'); return; else load image.mat P = [P,Image']; end %% %%载入验证数据 % load test.mat % P = [P,test_data']; %% 特征数据归一化 [P_scale,PS] = mapminmax(P); clear P %% 打乱排序，构建训练和测试样本 [trainV,valV,testV] = dividevec(P_scale(:,1:train_num),T,0.2,0);%训练数据和验证数据 testV.P = P_scale(:,train_num+1:end);%图像数据 % testV.T = test_label; % %% 建立LM-ANN神经网络 % % 建立网络结构 % net = newff(trainV.P,trainV.T,30,{'logsig' 'logsig'},'trainlm'); % % 设置训练参数 % % net.trainFcn = 'trainlm'; %'trainlm' --- Levenberg-Marquardt算法 'trainbfg'--- Quasi-Newton Algorithms-BFGS Algorithm % net.performFcn = 'sse'; %Sum squared error performance function % % net.performFcn = 'mse'; %Mean squared error performance function % net.trainParam.show = 1; %迭代输出间隔no. epochs between showing error % net.trainParam.lr = 0.05; %学习步长yita learning rate, not default trainlm % % net.trainParam.epochs = 1000; %最高训练次数n Max no. of epochs to train % net.trainParam.goal = 0.01; %目标精度e stop training if the error goal hit %% 循环试验模型精度选取精度较高的模型 Accuracy = 0.0; while (Accuracy < 0.85) %% 训练神经网络 % [net,tr] = train(net,trainV.P,trainV.T); %% 训练RBF-ANN神经网络 Q = size(trainV.P,2); % [net,tr] = newrb(trainV.P,trainV.T,0.01,0.9,Q,1); %Design radial basis network [net,tr] = newrb(trainV.P,trainV.T,0.01); %% 数据预测 T_predicted = sim(net,valV.P); %% 预测结果化整 [Max Index] = max(T_predicted); for j = 1:size(T_predicted,2) T_predicted(:,j) = 0; T_predicted(Index(j),j) = 1; end %% 准确率计算 delta = any(valV.T - T_predicted); acc_num = 0; for i=1:length(delta) if delta(i)==0 acc_num = acc_num + 1; end end Accuracy = acc_num/length(delta) end %% 保存网络 % save('net.mat','net'); %% 中间变量释放 clear Image P_scale T %% 对整幅图像进行预测 T_predicted=sim(net,testV.P); % test1 = testV.P(:,1:100000); % T_predicted1 = sim(net,test1); % test2 = testV.P(:,100001:200000); % T_predicted2 = sim(net,test2); % test3 = testV.P(:,200001:300000); % T_predicted3 = sim(net,test3); % test4 = testV.P(:,300001:363699); % T_predicted4 = sim(net,test4); % T_predicted=[T_predicted1,T_predicted2,T_predicted3,T_predicted4]; %% 预测结果化整 [Max Index] = max(T_predicted); for j = 1:size(T_predicted,2) T_predicted(:,j) = 0; T_predicted(Index(j),j) = 1; end %% 获得分类结果图像数据 pn2(1:line,1:sample,1:3)=0; I = zeros(size(T_predicted,2),1); image = zeros(line,sample); for i=1:size(T_predicted,2) I(i) = find(T_predicted(:,i)==1); end for i=1:line for j=1:sample image(i,j)=I((i-1)*line+j); pn2(i,j,1)=colorset(image(i,j),1); pn2(i,j,2)=colorset(image(i,j),2); pn2(i,j,3)=colorset(image(i,j),3); end end imshow(pn2); imwrite(pn2,'F:\output.tif'); % dlmwrite('RBF-ANN_asc.txt',image);%, '-append','delimiter', '\t', 'precision', '%10.6f','newline', 'pc'