Question

MATLAB CODE FOR

E xtreme learning machine using for classification task.

image processing electrical.

if you know then try or leave for other

Answer 1

clc;
clear;
close all;
II=imread('lena.tiff');
% I2=rgb2gray(II);
I2=II(:,:,3);
% figure;
% imshow(I2)
I=double(I2);
% I=I2;
%step1
LH{1}=I;
for i=2:5
[LA{i},LH{i},LV{i},LD{i}] = lwt2(LH{i-1},'haar');
end
LL4=LA{5};
C(:,1)=(reshape(LL4,1,1024))';
C(:,2)=(round(reshape(LL4,1,1024)/32))';
function [output_wt,TrainingAccuracy,TestingAccuracy,correctClassifiedSamples_Testing]=my_elm(training_data,testing_data, output_train,output_test,numberHidden)
%%Input arguments

function [output_wt,TrainingAccuracy,TestingAccuracy,correctClassifiedSamples_Testing]=my_elm(training_data,testing_data, output_train,output_test,numberHidden)
%%Input arguments
%training_data=N1*d (N1=# of training samples, N2=# of testing samples)
%output_train=N1*c (c= # of clases, d= feature dimension)
%testing_data=N2*d
%output_test=N2*c
start_time_train=cputime;
d=size(training_data,2);
training_data(:,d+1)=1; % input bias added in last column
%% input weight and bias (last column) initialization
inputWeightBias=rand(size(training_data,2),numberHidden)*0.0001;
%inputWeightBias=rand(size(training_data,2),numberHidden)*2-1; %a in [-1,1]
aa=inputWeightBias;
%a=rand(size(training_data,2),numberHidden)*2-1; %a in [-1,1]
H_final=[];
for i=1:size(training_data,1)
    h=inputWeightBias'*training_data(i,:)';
    H = 1 ./ (1 + exp(-h));% sigmoid trans func
    %H=h;
    %H=sin(h); % sine tf
    %H=radbas(h); % Radial basis function tf
    H_final=[H_final;H'];
    clear H;

end
output_wt=pinv(H_final)*output_train; % calculated output weight (beta)
end_time_train=cputime;
TrainingTime=end_time_train-start_time_train
%% Calculate the training accuracy
actual_output=H_final*output_wt; %the calculated output of the training data
correctClassifiedSamples_Training=0;
for i = 1 : size(output_train, 1)
        [x, index_desired_label]=max(output_train(i,:));
        [x, index_actual_label]=max(actual_output(i,:));
        if index_actual_label==index_desired_label
            correctClassifiedSamples_Training=correctClassifiedSamples_Training+1;
        end
end
    TrainingAccuracy=correctClassifiedSamples_Training/size(output_train,1)
%% testing part
start_time_test=cputime;
d1=size(testing_data,2);
testing_data(:,d1+1)=1; % input bias added in last column
H_final_test=[];
for i=1:size(testing_data,1)
    h_test=aa'*testing_data(i,:)';
    H_test = 1 ./ (1 + exp(-h_test));% sigmoid trans func
    %H_test=h_test;
    %H_test=sin(h_test); % sine tf
    %H_test=radbas(h_test); % Radial basis function tf
    H_final_test=[H_final_test;H_test'];
    clear H_test;
end
%% Calculate the testing accuracy
actual_output_test=H_final_test*output_wt; %the calculated output of the testing data
end_time_test=cputime;
TestingTime=end_time_test-start_time_test

correctClassifiedSamples_Testing=0;
for i = 1 : size(output_test, 1)
        [x, index_desired_label]=max(output_test(i,:));
        [x, index_actual_label]=max(actual_output_test(i,:));
        if index_actual_label==index_desired_label
            correctClassifiedSamples_Testing=correctClassifiedSamples_Testing+1;
        end
end
    TestingAccuracy=correctClassifiedSamples_Testing/size(output_test,1)
end

if you have any doubt feel free to ask in comments.