Add files via upload

Author: GSSfearless

SOM/SOM_net.m

@@ -0,0 +1,230 @@
+
+%%
+%clc; clear all; close all;
+clear;clc;close all
+tic;
+%location 
+
+path_train='train.txt';
+path_test = 'test.txt';
+train_label = repmat(1,1,676)
+test_label = repmat(1,1,676)
+train_data = load(path_train);
+test_data = load(path_test);
+[data_row,data_clown]=size(train_data);
+
+
+%s1 = struct('train_data_1',train_data_1,'train_label_1',train_label_1,'test_data_1',test_data_1,'test_label_1',test_label_1)
+
+
+
+%SOM network m*n
+m=10;
+n=10;
+%Neuron som_sum
+som_sum=m*n;
+%weight initialize 
+w = rand(som_sum, data_clown);
+%Initialize learning rate
+learn0 = 0.6;
+learn_rate = learn0;
+%learning para
+learn_para=100;
+%Setting iteration 
+iter = 50;%1000 500
+%Neuron location
+[I,J] = ind2sub([m, n], 1:som_sum);
+%Neighbor Initialize
+neighbor0 =2;
+neighbor_redius = neighbor0;
+%Neighbor parameters
+neighbor_para = 1000/log(neighbor0);
+
+for t=1:iter 
+    %  Scan all the sample points
+    display(['Iteration: ' num2str(t)]);
+    for j=1:data_row  
+        %Getting the winner neuron
+        data_x = train_data(j,:); 
+        %Find the winner neuron
+        [win_row, win_som_index]=min(dist(data_x,w'));  
+        %Find the winner neuron position
+        [win_som_row,win_som_cloumn] =  ind2sub([m, n],win_som_index);
+        win_som=[win_som_row,win_som_cloumn];
+        %Calculate and update the winner neuron neighbourhood position 
+        %distance_som = sum(( ([I( : ), J( : )] - repmat(win_som, som_sum,1)) .^2) ,2);
+        distance_som = exp( sum(( ([I( : ), J( : )] - repmat(win_som, som_sum,1)) .^2) ,2)/(-2*neighbor_redius*neighbor_redius)) ;
+        %Update the weights
+        for i = 1:som_sum
+           % if distance_som(i)<neighbor_redius*neighbor_redius 
+            w(i,:) = w(i,:) + learn_rate.*distance_som(i).*( data_x - w(i,:));
+        end
+    end
+ 
+    %Updateing the learning rate
+    learn_rate = learn0 * exp(-t/learn_para);   
+    %Update the neighbor radius
+    neighbor_redius = neighbor0*exp(-t/neighbor_para);  
+end
+
+
+
+%save as som_num
+som_num=cell(1,size(w,1));
+for i=1:size(w,1)
+    som_num{1,i}=[];
+end
+%Every neuron data corresponding number
+for num=1:data_row
+    [som_row,clown]= min(sum(( (w - repmat(train_data(num,:), som_sum,1)) .^2) ,2));
+    som_num{1,clown}= [som_num{1,clown},num];    
+end
+
+figure()
+for i=1:100
+subplot(10,10,i);
+weight=reshape(w(i,:),26,26);
+imshow(double(weight));
+end
+
+path1='som_num_1.mat';
+save(path1,'som_num');
+toc;
+%%
+number1 = load('p_dataset_26/Sample1/img002-00001.mat')
+number2 = load('p_dataset_26/Sample2/img003-00001.mat')
+number3 = load('p_dataset_26/Sample3/img004-00001.mat')
+numbera = load('p_dataset_26/SampleA/img011-00001.mat')
+numberb = load('p_dataset_26/SampleB/img012-00001.mat')
+numberc = load('p_dataset_26/SampleC/img013-00001.mat')
+%%
+fid = fopen('charact1.txt');
+% read a char at a time, ignore linefeed and carriage return
+% and put them in a 64 X 64 matrix
+lf = char(10); % line feed character
+cr = char(13); % carriage return character
+A = fscanf(fid, [cr lf '%c'],[64,64]);
+% close the file handler
+fclose(fid);
+A = A'; % transpose since fscanf returns column vectors
+% convert letters A‐V to their corresponding values in 32 gray levels
+% literal A becomes number 10 and so on...
+A(isletter(A))= A(isletter(A)) - 55;
+%convert number literals 0‐9 to their corresponding values in 32 gray
+%levels. Numeric literal '0' becomes number 0 and so on...
+A(A >= '0' & A <= '9') = A(A >= '0' & A <= '9') - 48;
+A = uint8(A);
+M = A
+
+% Preprocessing for test image
+Add_1 = zeros(26,2)
+N_1_o = M(1:26,1:24)
+N_1 = cat(2,Add_1,N_1_o)
+N_1_ad = imadjust(N_1,stretchlim(N_1),[])
+Nb1 = imcomplement(N_1_ad)
+N1 = reshape(Nb1,1,[])
+
+
+N_2 = M(1:26, 22:47)
+% N_2(1:2,:)=[]
+% N_2(21:end,:)=[]
+N_2_ad = imadjust(N_2,stretchlim(N_2),[])
+Nb2 = imcomplement(N_2_ad)
+N2 = reshape(Nb2,1,[])
+
+Add_3 = zeros(26,4)
+N_3_o = M(1:26, 43:64)
+N_3 = cat(2,N_3_o,Add_3)
+N_3(1:5,:)=[]
+N_3(15:21,:)=[]
+N_3_up = repmat(N_3(5,:),7,1)
+N_3_low = repmat(N_3(9,:),7,1)
+N_3_new = zeros(26,26)
+N_3_new(1:4,:) = N_3(1:4,:)
+N_3_new(5:11,:) = N_3_up
+N_3_new(12:15,:) = N_3(6:9,:)
+N_3_new(16:22,:) =N_3_low
+N_3_new(23:26,:) = N_3(11:14,:)
+%N_3_ad = imadjust(N_new,stretchlim(N_new),[])
+N_3_ad = imadjust(N_3_new,[0.01 0.02],[0 1])
+Nb3 = imcomplement(N_3_ad)
+Nb3 = Nb3.*255
+N3 = reshape(Nb3,1,[])
+
+Add_a = zeros(26,2)
+N_A_o = M(38:63, 1:24)
+N_A = cat(2,Add_a,N_A_o)
+N_a_ad = imadjust(N_A,stretchlim(N_A),[])
+Na = imcomplement(N_a_ad)
+NA = reshape(Na,1,[])
+
+
+N_B = M(38:63, 20:45)
+N_B(1,:)=[]
+N_B(21:25,:)=[]
+N_B_up= repmat(N_B(4,:),5,1)
+N_B_low= repmat(N_B(11,:),5,1)
+N_B_new = zeros(26,26)
+N_B_new(1:3,:)=N_B(1:3,:)
+N_B_new(4:6,:)= N_B(4:6,:)
+N_B_new(7:11,:)= N_B_up
+N_B_new(12:15,:)=N_B(9:12,:)
+N_B_new(16:20,:)=N_B_low
+N_B_new(21:26,:)=N_B(15:20,:)
+N_b_ad = imadjust(N_B_new,[0.01 0.02],[0 1])
+Nb = imcomplement(N_b_ad)
+NB = reshape(Nb,1,[])
+
+Add_C= zeros(26,1)
+N_C = M(38:63, 40:64)
+N_C = cat(2,N_C,Add_C)
+N_C(1,:)=[]
+N_C(21:end,:)=[]
+N_C_mid = repmat(N_C(10,:),11,1)
+N_C_new= zeros(26,26)
+N_C_new(1:7,:)=N_C(1:7,:)
+N_C_new(8:18,:)=N_C_mid
+N_C_new(19:26,:)=N_C(13:20,:)
+N_C_new(:,19:23)=zeros(26,5)
+N_C_ad = imadjust(N_C_new,[0.01 0.02],[0 1])
+Nc = imcomplement(N_C_ad)
+Nc = Nc.*255
+NC = reshape(Nc,1,[])
+
+n1 = load('train_1.txt')
+n2 = load('train_2.txt')
+n3 = load('train_3.txt')
+na = load('train_A.txt')
+nb = load('train_B.txt')
+nc = load('train_C.txt')
+n1_t = load('test_1.txt')
+n2_t = load('test_2.txt')
+n3_t = load('test_3.txt')
+na_t = load('test_A.txt')
+nb_t = load('test_B.txt')
+nc_t = load('test_C.txt')
+n = [n1;n2;n3;na;nb;nc]
+n_t = [n1_t; n2_t; n3_t; na_t; nb_t; nc_t]
+inputs = n';
+dimension1 = 10;
+dimension2 = 10;
+net = selforgmap([dimension1 dimension2]);
+[net,tr] = train(net,inputs);
+view(net)
+y = net(inputs);
+
+%t=mapminmax(t);
+% sim( ) to do Network simulation
+% r=sim(net,t);
+% % 
+% rr=vec2ind(r)
+% Find the topology structure of the network
+plotsomtop(net)
+% plot the distance plot
+plotsomnd(net)
+% plot the classifier result
+plotsomhits(net,inputs)
+
+
+
+