imgsimca.m

PLS_Toolbox是用于故障检测与诊断方面的matlab工具箱

function model = imgsimca(img,data)
%IMGSIMCA SIMCA Classification for Multivariate Images
%  This function allows the user to build SIMCA class models
%  by outlining areas on a reference image (img) that correspond
%  to areas on a multivariate image (data). This image must
%  be the same size (pixels by pixels) as the multivariate
%  image data but can have a different depth (possibly due to
%  having preprocessed the original data with IMGPCA). The
%  output is a structured array (model) which may be applied to 
%  new images with ISIMCAPR. The elements of the model are:
%
%          imgname: name of original image variable
%          mwaname: name of original multivariate image data
%             name: model type, always IMGSIMCA
%             date: date stamp
%             time: time stamp
%             size: size of the image data
%         simcamod: simca model parameters (see SIMCA)
%         in_class: array which gives pixels within 95% Q limits for each class
%    closest_class: array which gives closest class for each pixel
%            rqmat: reduced Q values for all classes for all pixels
%          rtsqmat: reduced T^2 values for all classes for all pixels
%
%I/O: model = imgsimca(img,data);
%
%See also IMGSELCT, IMGPCA, ISIMCAPR

%Copyright Eigenvector Research, Inc. 1999-2000
%bmw April 27, 1999
%nbg 11/00 added ;

h0 = image(img);    %nbg 11/00
title('Input image for selecting class areas')
datadim = size(data);
flag = 0;
k = 0;
while flag == 0
  disp('select an area')
  [xinds,yinds] = imgselct;
  k = k+1;
  cn = input('what class number is this? ');
  xlength = length(xinds);
  if k > 1;
    inds = [inds; [xinds yinds cn*ones(xlength,1)]];
  else
    inds = [xinds yinds cn*ones(xlength,1)];
  end
  nc = input('Do you want to do another area? ','s');
  if ~(nc(1) == 'y' | nc(1) == 'Y') 
    flag = 1;
  end
end
out = inds;
[mi,ni] = size(inds);
ndat = zeros(mi,datadim(3));
for i = 1:mi
  ndat(i,:) = squeeze(data(inds(i,2),inds(i,1),:))';
end
model = simca(ndat,inds(:,3));
h = get(0,'children');
close(h(1:2));
data = reshape(data,datadim(1)*datadim(2),datadim(3));
[nclass,rtsq,rq] = simcaprd(double(data),model,0);
[mr,nr] = size(rtsq);
classclose = ones(datadim(1),datadim(2),nr)*255;
classin = ones(datadim(1),datadim(2),nr+1)*255;
rqmat = zeros(datadim(1),datadim(2),nr);
rtsqmat = rqmat;
for i = 1:nr
  rqmat(:,:,i) = reshape(rq(:,i),datadim(1),datadim(2));
  rtsqmat(:,:,i) = reshape(rtsq(:,i),datadim(1),datadim(2));
end
k = 0;
for i = 1:datadim(2);
  for j = 1:datadim(1);
    k = k+1;
    classclose(j,i,nclass(k)) = 0;
	if ~isempty(find(rq(k,:)<=1))
	  classin(j,i,find(rq(k,:)<=1)) = zeros(size(find(rq(k,:)<=1))); 
    end
	if min(rq(k,:))>1
	  classin(j,i,nr+1) = 0;
	end
  end
end
colormap(hot(256));
for i = 1:nr
  if i == 1
    h1 = figure;
	image(classin(:,:,i)); 
	title('Areas in Class 1 Shown in Black')
    colormap(hot(256));
	h2 = figure;
	image(classclose(:,:,i));
	title('Areas Closest to Class 1 Shown in Black')
	colormap(hot(256))
	h3 = figure;
    imagesc(sqrt(rqmat(:,:,i)));
	title('Q values on Class 1 for Entire Image, 95% Limit = 1')
  else
    figure(h1); image(classin(:,:,i));
	title(sprintf('Areas in Class %g Shown in Black',i));
    figure(h2); image(classclose(:,:,i));
    title(sprintf('Areas Closest to Class %g Shown in Black',i));	
    figure(h3); imagesc(sqrt(rqmat(:,:,i)));
	title(sprintf('Q values on Class %g for Entire Image, 95% Limit = 1',i))
  end
  colormap(hot(256));
  colorbar
  pause
end
figure(h1); image(classin(:,:,nr+1));
title('Areas in Not in Any Class Shown in Black');
% Save the model as a structured array    
model = struct('imgname',inputname(1),'mwaname',inputname(2),'name','IMGSIMCA',...
        'date',date,'time',clock,'size',datadim,'simcamod',model,...
		'in_class',classin,'closest_class',classclose,'rqmat',rqmat,'rtsqmat',rtsqmat);