mltest1.m

传感器网络探测声信号通过特征提取进行目标设别分类

function [class,Distan]=MLtest1(Pt,Tt,Cen,cinv,ppri,mode)
% Usage: [class,conf]=MLtest1(Pt,Tt,Cen,cinv,ppri,mode)
% MLtest - Maximum Likelihood Classifier testing routing
% Pt: testing set feature vector  Q by N
% Tt: testing set target vector Q by S
% S: number of classes
% N: feature vector dimension
% K: # of feature vectors in training set
% Q: # of feature vectors in testing/validate set
% Cen: centroid of each cluster
% cinv: an array of inverse of covariance matrix of each class
% ppri: a priori probability = fraction of samples in each class
% class: classification result Q by S
% conf: 1 x Q vector. 
%       confidence = probability x is in class i given the observation vector Pt
% copyright (c) 2001  Yu Hen Hu
% Last revision 8/10/2001

[Q,N]=size(Pt);
ptnorm=ones(Q,1)./sqrt(sum(Pt'.*Pt')'); % Q X 1 vector
Pt = (ptnorm*ones(1,N)).*Pt; % make each row unity norm
[Q,S]=size(Tt);
if nargin<=4,
    ppri=ones(1,S)/S;
end
if S==1, % if only one output with 0 in one class 
         %   and 1 the other, change it to 2 outputs
  Tt=[Tt ones(Q,1)-Tt];
  S=2;
end

% Next, calculating the distance between test vector and i-th class Gaussian
%  using (X-Cen(cn,:))*C{i}^-1*(X-Cen(cn,:))  this is the negative log likelihood
for cn=1:S,
   if mode==0,
      tmpd=(Pt-ones(Q,1)*Cen(cn,:));    % Q x N
   elseif mode==1 | mode ==2,
      tmpd=Pt;
   end
  tmpe=tmpd*cinv{cn};               % Q x N
  Distan(cn,:) = sum(tmpe'.*tmpd'); % 1 X Q
end  
% Distan is S x Q matrix. 
% Now incorporating the a priori probability to give a MAP classifier
if mode==1|mode==0,
   Distan = Distan-diag(diag(ppri))*ones(1,Q);
end
[tmp,ind]=min(Distan);
% now calculate the confidence of the classification result.
% the confidence is calculated as the a posterior probability of the decision
% 
% edist=exp(-Distan);  % S x Q matrix containing est. of posterior prob of each class% edsum=ones(1,Q)./sum(edist);    % 1 x Q sum of the posterior prob.
% ppmat=ones(S,1)*edsum.*edist; % matrix of posterior probability

%Now try to find the maximum posterior probability 
% 

%[conf,ind]=max(ppmat); % given max element and index

%  Now test with Pt, Tt

Idmat=eye(S);
class=Idmat(ind,:);   % Q x S