cifar10数据集测试有多少张图_CIFAR-10数据集图像分类【PCA+基于最小错误率的贝叶斯决策】...

#CIFAR-10数据集：包含60000个32*32的彩色图像，共10类，每类6000个彩色图像。有50000个训练图像和10000个测试图像。

importscipy.io

train_data=scipy.io.loadmat("F:\\模式识别\\最小错误率的贝叶斯决策进行图像分类\\data\\train_data.mat")print(type(train_data))print(train_data.keys())print(train_data.values())print (len(train_data['Data']))#单张图片的数据向量长度：32X32X3=3072#内存占用量=3072*4*9968=116M 假定一个整数占用4个字节

print (len(train_data['Data'][0]))print(train_data)

x= train_data['Data']

y= train_data['Label']print(y)print(len(y))print(y.shape)print(y.flatten().shape)#labels_name:共5个标签，分别为airplane、automobile、bird、cat、deer

importscipy.io

labels_name=scipy.io.loadmat("F:\\模式识别\\最小错误率的贝叶斯决策进行图像分类\\data\\labels_name.mat")print(type(labels_name))print(labels_name)print(len(labels_name))#test_data：共5000个图像，5类，每类1000个图像

importscipy.io

test_data=scipy.io.loadmat("F:\\模式识别\\最小错误率的贝叶斯决策进行图像分类\\data\\test_data.mat")print (test_data['Label'])print (test_data['Data'])print (len(test_data['Label']))

datatest= test_data['Data']

labeltest= test_data['Label']print(datatest.shape)print(labeltest.shape)

test_index=[]for i inrange(len(labeltest)):if labeltest[i]==1:

test_index.append(i)elif labeltest[i]==2:

test_index.append(i)elif labeltest[i]==3:

test_index.append(i)#print (test_index)

labeltest=test_data['Label'][:3000]#print (labeltest)

importmatplotlib.pyplot as pltfrom mpl_toolkits.mplot3d importAxes3Dprint(x)print(x.shape)print(type(x))from sklearn.discriminant_analysis importLinearDiscriminantAnalysisfrom sklearn.decomposition importPCA

pca=PCA(n_components=0.79)#训练模型

pca.fit(x)

x_new=pca.transform(x)print("降维后各主成分的累计方差贡献率：",pca.explained_variance_ratio_)print("降维后主成分的个数：",pca.n_components_)print(x_new)

index_1=[]

index_2=[]

index_3=[]

index_num=[]for i inrange(len(y)):if y[i]==1:

index_1.append(i)elif y[i]==2:

index_2.append(i)elif y[i]==3:

index_3.append(i)

index_num=[len(index_1),len(index_2),len(index_3)]print(len(index_1))print(len(index_2))print(len(index_3))print(index_num)importnumpy as np

class1_feature=[]

class2_feature=[]

class3_feature=[]#index_1

for i inindex_1:

class1_feature.append(x_new[i])print(len(class1_feature))for i inindex_2:

class2_feature.append(x_new[i])print(len(class2_feature))for i inindex_3:

class3_feature.append(x_new[i])print(len(class3_feature))#计算第一类的类条件概率密度函数的参数

class1_feature=np.mat(class1_feature)print(class1_feature.shape)

miu1=[]

sigma1=[]for i in range(30):

miu=class1_feature[:,i].sum()/len(index_1)

miu1.append(miu)

temp=class1_feature[:,i]-miu

class1_feature[:,i]=temp

sigma1=(class1_feature.T*class1_feature)/len(index_1)print(miu1)print(sigma1)print(sigma1.shape)#计算第二类类条件概率密度函数的参数

class2_feature=np.mat(class2_feature)

miu2=[]

sigma2=[]for i in range(30):

miu=class2_feature[:,i].sum()/len(index_2)

miu2.append(miu)

temp=class2_feature[:,i]-miu

class2_feature[:,i]=temp

sigma2=(class2_feature.T*class2_feature)/len(index_2)print(miu2)print(sigma2)print(sigma2.shape)#计算第三类类条件概率密度函数的参数

class3_feature=np.mat(class3_feature)

miu3=[]

sigma3=[]for i in range(30):

miu=class3_feature[:,i].sum()/len(index_3)

miu3.append(miu)

temp=class3_feature[:,i]-miu

class3_feature[:,i]=temp

sigma3=(class3_feature.T*class3_feature)/len(index_3)print(miu3)print(sigma3)print(sigma3.shape)#计算三个类别的先验概率：

prior_index1=len(index_1)/len(y)

prior_index2=len(index_2)/len(y)

prior_index3=len(index_3)/len(y)print(prior_index1)print(prior_index2)print(prior_index3)importmath#降维

x_test =pca.transform(datatest)print(x_test)print(x_test.shape)print(x_test[0])#print ((np.mat(x_test[0]-miu1))*sigma1.I*(np.mat(x_test[0]-miu1).T))#print (((np.mat(x_test[0]-miu1))*sigma1.I*(np.mat(x_test[0]-miu1).T))[0,0])

predict_label=[]for i in range(3000):

g1=-0.5*((np.mat(x_test[i]-miu1))*sigma1.I*(np.mat(x_test[i]-miu1).T))[0,0]-0.5*math.log(np.linalg.det(sigma1))+math.log(prior_index1)

g2=-0.5*((np.mat(x_test[i]-miu2))*sigma2.I*(np.mat(x_test[i]-miu2).T))[0,0]-0.5*math.log(np.linalg.det(sigma2))+math.log(prior_index2)

g3=-0.5*((np.mat(x_test[i]-miu3))*sigma3.I*(np.mat(x_test[i]-miu3).T))[0,0]-0.5*math.log(np.linalg.det(sigma3))+math.log(prior_index3)if g1>g2:

max=1

if g1>g3:

max=1

else:

max=3

else:

max=2

if g2>g3:

max=2

else:

max=3predict_label.append(max)from sklearn.metrics importaccuracy_scoreprint (accuracy_score(predict_label,labeltest))