构建Alextenet 解决 Cifar100分类问题（1024）

import tensorflow as tf

# TODO  '注意'  如果下载数据失败则加入下面注释的两行代码
# import ssl
# ssl._create_default_https_context = ssl._create_unverified_context

from tensorflow import keras

cifar100 =keras.datasets.cifar100
(train_images, train_labels), _ = cifar100.load_data()   #下载数据
train_images = train_images/255.0       #归一化
train_images = tf.image.resize(train_images[:5000],(224,224))   #取前5000张照片并强制转换为（224*224*3）
train_labels = train_labels[:5000]  #取前5000张训练照片标签


# TODO.# ALEXNET解决Cifar100分类问题
#  TODO 1.conv1:11*11,96,4,valid   例：11*11（卷积核） ，stride=4(步长)
#   Maxpooling：3*3，2

# TODO 2.conv2：5*5，256，1，same
#  Maxpooling:3*3,2,valid

# TODO 3.  conv3：3*3，384，1，same

# TODO 4.  conv4：3*3，384，1，same
# TODO 5.  conv5：3*3，256，1，same
#  Maxpooling(最大池化):3*3,2,valid

#TODO 6.全连接层（打平） fc1:4096,dropout=0.5
# fc2:4096,dropout=0.5
# output:100（分为100类）
alexNet = keras.Sequential(layers=[

    #Conv1
    keras.layers.Conv2D(96,11,strides=(4,4),padding='valid',activation='relu'),
    keras.layers.BatchNormalization(),
    keras.layers.MaxPool2D(pool_size=(3,3),strides=(2,2),padding='valid'),
    #Conv2
    keras.layers.Conv2D(256,5,strides=(1,1),padding='same',activation='relu'),
    keras.layers.BatchNormalization(),
    keras.layers.MaxPool2D(pool_size=(3,3),strides=(2,2),padding='valid'),
    #conv3
    keras.layers.Conv2D(384,3,strides=(1*1),padding='same',activation='relu'),
    #Conv4
    keras.layers.Conv2D(384,3,strides=(1,1),padding='same',activation='relu'),
    #Conv5
    keras.layers.Conv2D(256,3,strides=(1,1),padding='same',activation='relu'),
    keras.layers.MaxPool2D(pool_size=(3,3),strides=(2,2),padding='valid'),
    #Flatten
    keras.layers.Flatten(),
    #FC1
    keras.layers.Dense(4096,activation='relu'),
    keras.layers.Dropout(rate=0.5),
    #fc2
    keras.layers.Dense(4096, activation='relu'),
    keras.layers.Dropout(rate=0.5),
    keras.layers.Dense(100)
])
#配置模型
alexNet.compile(optimizer='adam',
                loss=keras.losses.SparseCategoricalCrossentropy(from_logits=True),
                metrics=['accuracy']
                )

#训练模型
alexNet.fit(x=train_images,y=train_labels,batch_size=16,epochs=10)

如有侵权，请联系我删除