project4.asv

模式识别一份很好的作业,包括线性分类器；最小风险贝叶斯分类器；监督学习法分层聚类分析；K－L变换提取有效特征,支持向量机

%function U=select_feature(sel)
sel=120;
load human;
load vehicle;
k_s=1;
k_d=120;
hm_tr(250,252)=0;
hm_test(250,252)=0;
[hm_tr,hm_test]=randQ(human);
randQ(vehicle);
bg_tr(250,252)=0;
bg_test(250,252)=0;
[veh_tr,veh_test]=randQ(vehicle);  
M_hm=hm_tr(1,:);
for i=2:250
M_hm=M_hm+(hm_tr(i));
end
M_hm=M_hm/250;
M_veh=veh_tr(1,:);
for i=2:250
M_veh=M_veh+(veh_tr(i));
end
M_veh=M_veh/250;
E_hm(252,252)=0;
E_veh(252,252)=0;
for i=1:250
    E_hm=E_hm+(hm_tr(i,:)-M_hm)'*(hm_tr(i,:)-M_hm);
    E_veh=E_veh+(veh_tr(i,:)-M_veh)'*(veh_tr(i,:)-M_veh);
end
E_hm=E_hm/250;
E_veh=E_veh/250;
%假设行人和骑车两类的先验概率都相等，均等于0.5;
Sw=0.5*E_hm+0.5+E_veh;
[V,D]=eig(Sw);
M_S(1,252)=0;
for i=1:250
M_S=M_S+hm_tr(i)+veh_tr(i);
end
Sb=0.5*(M_hm-M_S)'*(M_hm-M_S)+0.5*(M_veh-M_S)'*(M_veh-M_S);
Y(1,252)=0;
for i=1:252
    for j=1:252
        if (i==j)
            Y(i)=V(i,j);
        end
    end
end
for i=1:252
    J(i)=V(:,i)'*Sb*V(:,i)/Y(i);
end
index(1:252)=0;
for i=1:252
   [K,I]=max(J);
   index(i)=I;
   J(I)=-1.0e+19;
end
U(252,252)=0;
for i=1:252
    a=index(i);
    U(:,i)=V(:,a);
end
U_OK(252,sel)=0;
for i=1:sel
U_OK(:,i)=U(:,i);
end
hm_tr_kl=U_OK'*hm_tr';
veh_tr_kl=U_OK'*veh_tr';
hm_test_kl=U_OK'*hm_test';
veh_test_kl=U_OK'*veh_test';
Y(500,1)=0;
a(1,250)=0;
a(1,:)=1;
b(1,250)=0;
b(1,:)=-1;
c=[a,b]';
tr_kl=[hm_tr_kl,veh_tr_kl]';
test_kl=[hm_test_kl,veh_test_kl]';
net=svm(sel,'rbf',5,1,0,'quadprog',50);
K=svmkernel(net,hm_tr_kl',veh_tr_kl');
net=svmtrain(net,tr_kl,c);
Y=svmfwd(net,test_kl);