keras实现用resnet18非预训练模型对四种天气分类，能自己定义归一化方法，并绘制图像

好的，我们可以按照以下步骤来实现：

1. 导入必要的库和模块

from tensorflow import keras
from tensorflow.keras import layers
from tensorflow.keras.preprocessing.image import ImageDataGenerator
import matplotlib.pyplot as plt

2. 定义数据路径和超参数

data_dir = 'path/to/data'
img_height = 224
img_width = 224
batch_size = 32
num_epochs = 10

3. 定义归一化方法

# 自定义归一化方法
def normalize_img(img):
    img = tf.cast(img, dtype=tf.float32) / 255.
    return (img - 0.5) / 0.5

4. 加载数据并进行预处理

# 使用 ImageDataGenerator 加载数据
datagen = ImageDataGenerator(
    preprocessing_function=normalize_img
)

train_data = datagen.flow_from_directory(
    directory=data_dir + '/train',
    target_size=(img_height, img_width),
    batch_size=batch_size,
    class_mode='categorical',
)

val_data = datagen.flow_from_directory(
    directory=data_dir + '/val',
    target_size=(img_height, img_width),
    batch_size=batch_size,
    class_mode='categorical',
)

5. 构建模型

def resnet_block(inputs, filters, kernel_size, strides, activation='relu'):
    x = layers.Conv2D(filters, kernel_size=kernel_size, strides=strides, padding='same')(inputs)
    x = layers.BatchNormalization()(x)
    if activation:
        x = layers.Activation(activation)(x)
    return x

def resnet18(input_shape, num_classes):
    inputs = layers.Input(shape=input_shape)

    # 第一层卷积
    x = resnet_block(inputs, filters=64, kernel_size=7, strides=2)

    # 残差块
    x = layers.MaxPooling2D(pool_size=3, strides=2, padding='same')(x)
    for _ in range(2):
        x = resnet_block(x, filters=64, kernel_size=3, strides=1)
    x = layers.MaxPooling2D(pool_size=3, strides=2, padding='same')(x)
    for _ in range(2):
        x = resnet_block(x, filters=128, kernel_size=3, strides=1)
    x = layers.MaxPooling2D(pool_size=3, strides=2, padding='same')(x)
    for _ in range(2):
        x = resnet_block(x, filters=256, kernel_size=3, strides=1)
    x = layers.MaxPooling2D(pool_size=3, strides=2, padding='same')(x)
    for _ in range(2):
        x = resnet_block(x, filters=512, kernel_size=3, strides=1)

    # 全局平均池化
    x = layers.GlobalAveragePooling2D()(x)

    # 输出层
    outputs = layers.Dense(num_classes, activation='softmax')(x)

    # 模型实例化
    model = keras.Model(inputs, outputs, name='resnet18')
    return model

# 实例化模型
model = resnet18(input_shape=(img_height, img_width, 3), num_classes=4)

6. 编译模型

model.compile(
    optimizer=keras.optimizers.Adam(),
    loss=keras.losses.CategoricalCrossentropy(),
    metrics=[keras.metrics.CategoricalAccuracy()],
)

7. 训练模型

history = model.fit(
    train_data,
    epochs=num_epochs,
    validation_data=val_data,
)

8. 绘制准确率和损失函数变化图

plt.plot(history.history['categorical_accuracy'], label='train')
plt.plot(history.history['val_categorical_accuracy'], label='val')
plt.title('Accuracy')
plt.xlabel('Epoch')
plt.ylabel('Accuracy')
plt.legend()
plt.show()

plt.plot(history.history['loss'], label='train')
plt.plot(history.history['val_loss'], label='val')
plt.title('Loss')
plt.xlabel('Epoch')
plt.ylabel('Loss')
plt.legend()
plt.show()

这样就完成了用resnet18非预训练模型对四种天气分类的实现，并绘制了准确率和损失函数变化图。