卷积神经网络__基于cifar100数据集

1.数据加载及预处理

# 数据预处理  将数据进行类型转换
def process(x, y):
    x = tf.cast(x, dtype=tf.float32) / 255.
    y = tf.cast(y, dtype=tf.int64)
    return x, y

# 加载数据 并进行 shuffle 和 batch 等操作
(x, y), (x_test, y_test) = datasets.cifar100.load_data()  
y = tf.squeeze(y, axis=1)
y_test = tf.squeeze(y_test, axis=1)

db_train = tf.data.Dataset.from_tensor_slices((x, y))
db_train = db_train.shuffle(1000).map(process).batch(64)

db_test = tf.data.Dataset.from_tensor_slices((x, y))
db_test = db_test.map(process).batch(64)

数据集有 160M , 用 pycharm 直接加载很慢很慢,建议去官网复制下载链接到迅雷,然后粘贴到 ‘’ C:\Users\SaltedFish.keras\datasets ‘’ 目录, 就直接可以用
官网下载链接: http://www.cs.toronto.edu/~kriz/cifar-100-python.tar.gz

2.建立卷积网络并训练、预测并的出准确率

conv_layers = [    
    layers.Conv2D(64, kernel_size=[3, 3], padding='same', activation=tf.nn.relu), # 卷积层 取的是局部特征, 通过局部特征来表示
    layers.Conv2D(64, kernel_size=[3, 3], padding='same', activation=tf.nn.relu),
    layers.MaxPool2D(pool_size=[2, 2], strides=2, padding='same'),# 池化层 保留主要的特征同时减少参数(降维，效果类似PCA)和计算量，防止过拟合，提高模型泛化能力

    layers.Conv2D(128, kernel_size=[3, 3], padding='same', activation=tf.nn.relu),
    layers.Conv2D(128, kernel_size=[3, 3], padding='same', activation=tf.nn.relu),
    layers.MaxPool2D(pool_size=[2, 2], strides=2, padding='same'),
    
    layers.Conv2D(256, kernel_size=[3, 3], padding='same', activation=tf.nn.relu),
    layers.Conv2D(256, kernel_size=[3, 3], padding='same', activation=tf.nn.relu),
    layers.MaxPool2D(pool_size=[2, 2], strides=2, padding='same'),
    
    layers.Conv2D(512, kernel_size=[3, 3], padding='same', activation=tf.nn.relu),
    layers.Conv2D(512, kernel_size=[3, 3], padding='same', activation=tf.nn.relu),
    layers.MaxPool2D(pool_size=[2, 2], strides=2, padding='same'),
    
    layers.Conv2D(512, kernel_size=[3, 3], padding='same', activation=tf.nn.relu),
    layers.Conv2D(512, kernel_size=[3, 3], padding='same', activation=tf.nn.relu),
    layers.MaxPool2D(pool_size=[2, 2], strides=2, padding='same'),
]
conv_net = Sequential(conv_layers)  # 建立神经网络
conv_net.build(input_shape=[None, 32, 32, 3])  # 设置传入的 shape
# 建立全连接层
fc_net = Sequential([
        layers.Dense(256, activation=tf.nn.relu),
        layers.Dense(128, activation=tf.nn.relu),
        layers.Dense(100, activation=None)
    ])
    # 设置全连接层的输入 shape
    fc_net.build(input_shape=[None, 512])
	# 设置优化器
    optimizer = optimizers.Adam(lr=1e-4)
	# 将两个网络的参数合成一个参数列表 进行更新
    variables = conv_net.trainable_variables + fc_net.trainable_variables

    acc_meter = metrics.Accuracy()
    loss_meter = metrics.Mean()

    for epoch in range(50):
        for step, (x, y) in enumerate(db_train):
            with tf.GradientTape() as tape:
                # [b,32,32,3] ==> [b,1,1,512]
                out = conv_net(x)
                out = tf.reshape(out, [-1, 512])
                # [b,512] ==> [b,100]
                logits = fc_net(out)
                y_one_hot = tf.one_hot(y, depth=100)
                loss = tf.losses.categorical_crossentropy(y_one_hot, logits, from_logits=True)
                # 更新 loss 函数
                loss_meter.update_state(loss)

            grads = tape.gradient(loss, conv_net.trainable_variables + fc_net.trainable_variables)
            optimizer.apply_gradients(zip(grads, variables))

            if step % 100 == 0:
                print(epoch, step, 'loss:', loss_meter.result().numpy())
                loss_meter.reset_states()

        for step, (x, y) in enumerate(db_test):
            with tf.GradientTape() as tape:
                out = conv_net(x)
                out = tf.reshape(out, [-1, 512])
                logits = fc_net(out)
                prob = tf.nn.softmax(logits, axis=1)
                pred = tf.argmax(prob, axis=1)
                acc_meter.update_state(y, pred)

        print(epoch, "acc result:", acc_meter.result().numpy())