基于宽度学习实现人脸识别附matlab代码.zip

function [TrainingAccuracy,TestingAccuracy,Training_time,Testing_time] = bls_train_bp(train_x,train_y,test_x,test_y,s,C,N1,N2,N3) % Learning Process of the proposed broad learning system trained Wh by BP % algorithms. Moreover, the number of epochs of the bp algorithm is set as % 50 by default. %Input: %---train_x,test_x : the training data and learning data %---train_y,test_y : the label %---We: the randomly generated coefficients of feature nodes %---wh:the randomly generated coefficients of enhancement nodes %----s: the shrinkage parameter for enhancement nodes %----C: the regularization parameter for sparse regualarization %----N11: the number of feature nodes per window %----N2: the number of windows of feature nodes %%%%%%%%%%%%%%feature nodes%%%%%%%%%%%%%% tic train_x = zscore(train_x')'; H1 = [train_x .1 * ones(size(train_x,1),1)];y=zeros(size(train_x,1),N2*N1); for i=1:N2 we=2*rand(size(train_x,2)+1,N1)-1; We{i}=we; A1 = H1 * we;A1 = mapminmax(A1); clear we; beta1 = sparse_bls(A1,H1,1e-3,50)'; beta11{i}=beta1; % clear A1; T1 = H1 * beta1; fprintf(1,'Feature nodes in window %f: Max Val of Output %f Min Val %f\n',i,max(T1(:)),min(T1(:))); [T1,ps1] = mapminmax(T1',0,1);T1 = T1'; ps(i)=ps1; % clear H1; % y=[y T1]; y(:,N1*(i-1)+1:N1*i)=T1; end clear H1; clear T1; %%%%%%%%%%%%%enhancement nodes%%%%%%%%%%%%%%%%%%%%%%%%%%%% H2 = [y .1 * ones(size(y,1),1)]; if N1*N2>=N3 wh=orth(2*rand(N2*N1+1,N3)-1); else wh=orth(2*rand(N2*N1+1,N3)'-1)'; end T2 = H2 * wh; l2 = max(max(T2)); l2 = s/l2; fprintf(1,'Enhancement nodes: Max Val of Output %f Min Val %f\n',l2,min(T2(:))); T2 = tansig(T2 * l2); T3=[y T2]; % clear H2;clear T2; beta = (T3' * T3+eye(size(T3',1)) * (C)) \ ( T3' * train_y); %%%%%%%%%%%%%%%back propogation for wh%%%%%%%%%%%%%%%%%%%%%%%%%%%%% xx = T3 * beta; yy = result(xx); train_yy = result(train_y); TrainingAccuracy = length(find(yy == train_yy))/size(train_yy,1); disp(['Training Accuracy is : ', num2str(TrainingAccuracy * 100), ' %' ]); test_x = zscore(test_x')'; HH1 = [test_x .1 * ones(size(test_x,1),1)]; %clear test_x; yy1=zeros(size(test_x,1),N2*N1); for i=1:N2 beta1=beta11{i};ps1=ps(i); TT1 = HH1 * beta1; TT1 = mapminmax('apply',TT1',ps1)'; clear beta1; clear ps1; %yy1=[yy1 TT1]; yy1(:,N1*(i-1)+1:N1*i)=TT1; end HH2 = [yy1 .1 * ones(size(yy1,1),1)]; TT2 = tansig(HH2 * wh * l2);TT3=[yy1 TT2]; clear TT2; x = TT3 * beta; y1 = result(x); test_yy = result(test_y); TestingAccuracy = length(find(y1 == test_yy))/size(test_yy,1); clear TT3; disp(['Testing Accuracy is : ', num2str(TestingAccuracy * 100), ' %' ]); m=1;N=size(y,1);wh1=ones(1,N);Accuracy_b=zeros(1,5); beta_b=zeros(size(beta,1),size(beta,2),5); T3_b=zeros(size(T3,1),size(T3,2),5); wh_b=zeros(size(wh,1),size(wh,2),5); l2_b=zeros(size(l2,1),size(l2,2),5); T2_b=zeros(size(T2,1),size(T2,2),5); while m<=50 xx = T3 * beta; W2=beta(N1*N2+1:end,:); Wh=wh(1:N1*N2,:); Wh_beta=wh(N1*N2+1,:); %T2temp=(T2.*(1-T2)); T2temp=((1-T2.^2)); tempWh=y'*(((xx-train_y)*W2').*T2temp); Wh=Wh+0.1*tempWh; %Wh=Wh+0.1*l2*tempWh; tempWhbeta=wh1*(((xx-train_y)*W2').*T2temp); Wh_beta=Wh_beta+0.1*tempWhbeta; %Wh_beta=Wh_beta+0.1*l2*tempWhbeta; wh_temp=[Wh; Wh_beta]; T2_temp = H2 * wh_temp; l2_temp = max(max(T2_temp)); l2_temp = s/l2_temp; fprintf(1,'Enhancement nodes in epochs %f : Max Val of Output %f Min Val %f\n',m,l2_temp,min(T2_temp(:))); T2_temp = tansig(T2_temp * l2_temp); T3_temp=[y T2_temp]; beta_temp = (T3_temp' * T3_temp+eye(size(T3_temp',1)) * (C)) \ ( T3_temp' * train_y); %%%%%%%%%%%%%%%%%%%%%%%%%%The following block is for the case that %%%%%%%%%%%%%%%%%%%%%%%%%%when the next 4 iterations' training accuracy %%%%%%%%%%%%%%%%%%%%%%%%%%is not better than the result of the mth %%%%%%%%%%%%%%%%%%%%%%%%%%iteration.%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% xx_temp = T3_temp * beta_temp; yy_temp = result(xx_temp); train_yy = result(train_y); TrainingAccuracy = length(find(yy_temp == train_yy))/size(train_yy,1); disp(['Training Accuracy is : ', num2str(TrainingAccuracy * 100), ' %' ]); Accuracy=TrainingAccuracy; %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%The following block is for the case that %%%%%%%%%%%%%%%%%%%%%%%%%%when the next 4 iterations' testing accuracy %%%%%%%%%%%%%%%%%%%%%%%%%%is not better than the result of the mth %%%%%%%%%%%%%%%%%%%%%%%%%%iteration.%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % TT2 = tansig(HH2 * wh_temp * l2_temp);TT3=[yy1 TT2]; % clear TT2; % x = TT3 * beta_temp; % y1 = result(x); % test_yy = result(test_y); % TestingAccuracy = length(find(y1 == test_yy))/size(test_yy,1); % clear TT3; % disp(['Testing Accuracy is : ', num2str(TestingAccuracy * 100), ' %' ]); % Accuracy=TestingAccuracy; %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% if m<=5 beta=beta_temp; beta_b(:,:,m)=beta_temp; T3=T3_temp;T3_b(:,:,m)=T3_temp; wh=wh_temp;wh_b(:,:,m)=wh_temp; l2=l2_temp;l2_b(:,:,m)=l2_temp; T2=T2_temp;T2_b(:,:,m)=T2_temp; Accuracy_b(m)=Accuracy; m=m+1; else beta_b(:,:,1:4)=beta_b(:,:,2:5); beta_b(:,:,5)=beta_temp; T3_b(:,:,1:4)=T3_b(:,:,2:5);T3_b(:,:,5)=T3_temp; wh_b(:,:,1:4)=wh_b(:,:,2:5);wh_b(:,:,5)=wh_temp; l2_b(:,:,1:4)=l2_b(:,:,2:5);l2_b(:,:,5)=l2_temp; T2_b(:,:,1:4)=T2_b(:,:,2:5);T2_b(:,:,5)=T2_temp; Accuracy_b(1:4)=Accuracy_b(2:5);Accuracy_b(5)=Accuracy; [a,index]=max(Accuracy_b); if index~=1 beta=beta_temp; T3=T3_temp; wh=wh_temp; l2=l2_temp; T2=T2_temp; m=m+1; else beta=beta_b(:,:,1); T3=T3_b(:,:,1); wh=wh_b(:,:,1); l2=l2_b(:,:,1); T2=T2_b(:,:,1); break end end end Training_time = toc; disp('Training has been finished!'); disp(['The Total Training Time is : ', num2str(Training_time), ' seconds' ]); %%%%%%%%%%%%%%%%%Training Accuracy%%%%%%%%%%%%%%%%%%%%%%%%%% xx = T3 * beta; clear T3; yy = result(xx); train_yy = result(train_y); TrainingAccuracy = length(find(yy == train_yy))/size(train_yy,1); disp(['Training Accuracy is : ', num2str(TrainingAccuracy * 100), ' %' ]); tic; %%%%%%%%%%%%%%%%%%%%%%Testing Process%%%%%%%%%%%%%%%%%%% % test_x = zscore(test_x')'; % HH1 = [test_x .1 * ones(size(test_x,1),1)]; % %clear test_x; % yy1=zeros(size(test_x,1),N2*N1); % for i=1:N2 % beta1=beta11{i};ps1=ps(i); % TT1 = HH1 * beta1; % TT1 = mapminmax('apply',TT1',ps1)'; % % clear beta1; clear ps1; % %yy1=[yy1 TT1]; % yy1(:,N1*(i-1)+1:N1*i)=TT1; % end % clear TT1;clear HH1; % HH2 = [yy1 .1 * ones(size(yy1,1),1)]; TT2 = tansig(HH2 * wh * l2);TT3=[yy1 TT2]; clear HH2;clear wh;clear TT2; %%%%%%%%%%%%%%%%% testing accuracy%%%%%%%%%%%%%%%%%%%%%%%%%%% x = TT3 * beta; y = result(x); test_yy = result(test_y); TestingAccuracy = length(find(y == test_yy))/size(test_yy,1); clear TT3; Testing_time = toc; disp('Testing has been finished!'); disp(['The Total Testing Time is : ', num2str(Testing_time), ' seconds' ]); disp(['Testing Accuracy is : ', num2str(TestingAccuracy * 100), ' %' ]);