CIFAR-10数据集： https://www.cs.toronto.edu/~kriz/cifar.html

1
2
3
4
5
6

import numpy as np
import pandas as pd
import keras
from keras import layers
import matplotlib.pyplot as plt
import keras.datasets.cifar10 as cifar

1. 准备数据：

1

(train_image,train_label),(test_image,test_label)=cifar.load_data()

1
2
3

#归一化处理
train_image=train_image/255
test_image = test_image/255

1
2

#查看数据
test_image.shape

(10000, 32, 32, 3)

1

plt.imshow(train_image[2])

<matplotlib.image.AxesImage at 0x26981d7c220>

1

train_label[2]

array([9], dtype=uint8)

1

**2. 构建网络模型：**

1
2

model = keras.Sequential()
model.add(layers.Conv2D())

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16

#step1 卷积神经网络
model.add(layers.Conv2D(64,(3,3),input_shape=(32,32,3),activation='relu')) #input layers
model.add(layers.Conv2D(64,(3,3),activation='relu')) #Convolution layers
model.add(layers.MaxPool2D()) #pooling
model.add(layers.Conv2D(64,(3,3),activation='relu'))
model.add(layers.Conv2D(64,(3,3),activation='relu'))
model.add(layers.MaxPool2D())
#step2 展平数据
model.add(layers.Flatten()) # 
#step3 构建全链接层
model.add(layers.Dense(256,activation='relu'))
model.add(layers.Dropout(0.5))
model.add(layers.Dense(256,activation='relu'))
model.add(layers.Dropout(0.5))
model.add(layers.Dense(10,activation='softmax'))

1

model.summary()

Model: "sequential"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 conv2d (Conv2D)             (None, 30, 30, 64)        1792      
                                                                 
 conv2d_1 (Conv2D)           (None, 28, 28, 64)        36928     
                                                                 
 max_pooling2d (MaxPooling2D  (None, 14, 14, 64)       0         
 )                                                               
                                                                 
 conv2d_2 (Conv2D)           (None, 12, 12, 64)        36928     
                                                                 
 conv2d_3 (Conv2D)           (None, 10, 10, 64)        36928     
                                                                 
 max_pooling2d_1 (MaxPooling  (None, 5, 5, 64)         0         
 2D)                                                             
                                                                 
 flatten (Flatten)           (None, 1600)              0         
                                                                 
 dense (Dense)               (None, 256)               409856    
                                                                 
 dropout (Dropout)           (None, 256)               0         
                                                                 
 dense_1 (Dense)             (None, 256)               65792     
                                                                 
 dropout_1 (Dropout)         (None, 256)               0         
                                                                 
 dense_2 (Dense)             (None, 10)                2570      
                                                                 
=================================================================
Total params: 590,794
Trainable params: 590,794
Non-trainable params: 0
_________________________________________________________________

3. 编译模型：

1

model.compile(optimizer='adam',loss='sparse_categorical_crossentropy',metrics=['acc'])

4. 训练模型：

1

model.fit(train_image,train_label,epochs=10,batch_size=512,validation_data=(test_image,test_label))

Epoch 1/10
98/98 [==============================] - 166s 2s/step - loss: 1.9493 - acc: 0.2648 - val_loss: 193.4561 - val_acc: 0.3232
Epoch 2/10
98/98 [==============================] - 170s 2s/step - loss: 1.5473 - acc: 0.4304 - val_loss: 148.5261 - val_acc: 0.4404
Epoch 3/10
98/98 [==============================] - 176s 2s/step - loss: 1.3480 - acc: 0.5138 - val_loss: 146.2150 - val_acc: 0.4511
Epoch 4/10
98/98 [==============================] - 168s 2s/step - loss: 1.2286 - acc: 0.5620 - val_loss: 200.3967 - val_acc: 0.4550
Epoch 5/10
98/98 [==============================] - 165s 2s/step - loss: 1.1045 - acc: 0.6099 - val_loss: 187.7144 - val_acc: 0.4452
Epoch 6/10
98/98 [==============================] - 164s 2s/step - loss: 1.0141 - acc: 0.6465 - val_loss: 208.9634 - val_acc: 0.4655
Epoch 7/10
98/98 [==============================] - 178s 2s/step - loss: 0.9519 - acc: 0.6722 - val_loss: 105.2857 - val_acc: 0.5780
Epoch 8/10
98/98 [==============================] - 171s 2s/step - loss: 0.8804 - acc: 0.6943 - val_loss: 176.2543 - val_acc: 0.4892
Epoch 9/10
98/98 [==============================] - 173s 2s/step - loss: 0.8249 - acc: 0.7135 - val_loss: 150.6743 - val_acc: 0.5347
Epoch 10/10
98/98 [==============================] - 174s 2s/step - loss: 0.7784 - acc: 0.7310 - val_loss: 111.6608 - val_acc: 0.5940





<keras.callbacks.History at 0x269ee9bd7c0>

5. 评估模型：

1

model.evaluate(test_image,test_label)

313/313 [==============================] - 19s 59ms/step - loss: 2.3031 - acc: 0.1072





[2.3030948638916016, 0.10719999670982361]

1

model.evaluate(train_image,train_label)

1563/1563 [==============================] - 77s 49ms/step - loss: 2.3031 - acc: 0.1059





[2.3030762672424316, 0.10586000233888626]

5. 预测数据：

1

pred_label = model.predict(test_image)

1

pred=np.argmax(pred_label[0])

1
2

if pred == 4:
    print("deer")

deer

1
2

plt.figure(figsize=(1,1)) #设置尺寸
plt.imshow(test_image[3])

<matplotlib.image.AxesImage at 0x269823c8280>

1

1